KR20150129839A - System-wide checkpoint avoidance for distributed database systems - Google Patents

Abstract

Distributed database systems can implement fast failover. Upon recovery from a database head node failure, a connection with one or more storage nodes of a distributed storage system that stores data for a database implemented by a database head node may be established. Upon establishing a connection with the storage nodes, the database may be made available for access, such as various access requests. In various embodiments, the redo log record may not be replayed to provide access to the database. In at least some embodiments, the storage nodes may provide the current state of the stored data for the database in response to the requests.

Description

{SYSTEM-WIDE CHECKPOINT AVOIDANCE FOR DISTRIBUTED DATABASE SYSTEMS}

The variance of the various components of the software stack may, in some cases, increase due to high tensions (e.g., through replication), high persistence, and low cost solutions (e.g., , Or through low-cost components). However, databases have historically been among the components of the software stack that are the most difficult to distribute. For example, it may be difficult to distribute databases while still ensuring the so-called ACID attributes (e.g., atomicity, consistency, separability, and persistence) they are expected to provide.

Most existing relational databases are not distributed, but some existing databases are "scaled out" using only one of two common models: the "non-shared" model and the "shared disk" model In contrast to being "scaled up" by employing a large monolithic system). Generally, in the "non-shared" model, the received queries are decomposed into database shards (each of which contains a component of the query) and these shards are sent to the different compute nodes And the results are collected and aggregated before they are returned. Generally, in the "shared disk" model, all compute nodes in the cluster access the same underlying data. In systems employing this model, care must be taken to manage cache coherence. In both of these models, a large, monolithic database is replicated on a number of nodes (including all the functionality of a stand-alone database instance) and "glue" logic is added to stitch them together. do. For example, in the "non-shared" model, glue logic can provide the functionality of a dispatcher that subdivides queries, sends them to multiple compute notes, and then combines the results. In the "shared disk" model, glue logic can serve to fuse caches of multiple nodes together (e.g., to manage consistency in the caching hierarchy). These "non-shared" and "shared disk" database systems can be expensive to deploy and complex to maintain, and can master many database use cases.

1 is a block diagram illustrating various components of a database software stack, in accordance with one embodiment.
2 is a block diagram illustrating a service system architecture that may be configured to implement a database service based on web services, in accordance with some embodiments.
3 is a block diagram illustrating various components of a database system including a database engine and an individual distributed database storage service, in accordance with one embodiment.
4 is a block diagram illustrating a distributed database-optimized storage system, according to one embodiment.
5 is a block diagram illustrating use of an individual distributed database-optimized storage system in a database system, in accordance with one embodiment.
6 is a block diagram illustrating how data and metadata may be stored at a given node of a distributed database-optimized storage system, in accordance with one embodiment.
7 is a block diagram illustrating an exemplary configuration of a database volume, according to one embodiment.
Figure 8 is a flow chart illustrating a method for avoiding checkpoints on the entire system in a distributed database system, in accordance with some embodiments.
Figure 9A is a set of examples illustrating a method for performing fast failover for a distributed database system, in accordance with some embodiments.
Figure 9B is a flow chart illustrating a method for performing fast failover for a distributed database system, in accordance with some embodiments.
9C is a flow diagram illustrating a method for processing access requests in a recovered database, in accordance with some embodiments.
10 is a block diagram illustrating a computer system configured to implement at least a portion of a database system including a database engine and an individual distributed database storage service, in accordance with various embodiments.
Although the embodiments are described in this application as examples for some embodiments and illustrative drawings, those of ordinary skill in the art will recognize that the embodiments are not limited to the embodiments or drawings described. The drawings and detailed description are not intended to limit the embodiments to the particular forms disclosed, but on the contrary, it is intended to cover all modifications, equivalents, and alternatives falling within the spirit and scope as defined by the appended claims. It is to be understood that it is intended to encompass water. The headings used in this application are not intended to be used to limit the scope of the present description or the claims, but merely for structural purposes. As used throughout this application, the word "can" is used in a generous meaning (i.e., a possible meaning), rather than an essential meaning (i.e. The words " comprising, "" including, "and" comprising "mean inclusive and include, but are not limited to, open relationships. Similarly, the words " having ","having",and" having "also mean open relationships and thus include, but are not limited to. The terms " first, "" second, "third," and the like as used in this application are to be construed in a manner similar to the order in which they appear, Spatial, temporal, logical, etc.).
The various components may be described as "configured to " perform tasks or tasks. In such contexts, "configured to" is a broad listing encompassing "having a structure" that generally performs tasks or tasks during operation. As such, an element may be configured to perform an operation even when the element is not currently performing the operation (e.g., the computer system is configured to perform operations even when no operations are currently being performed . In some contexts, "configured to" can generally be a broad listing of structures that mean "having a circuit " that performs tasks or tasks during operation. As such, a component can be configured to perform an operation even when the component is not currently in operation. In general, the circuitry forming the structure corresponding to "configured to" may comprise hardware circuits.
The various components may be described as performing tasks or tasks for convenience in the description. Such descriptions should be interpreted to include the phrase "configured to" Listing components that are configured to perform more than one task is not specifically intended to apply the 35 USC § 112, paragraph 6, interpretation for that component.
On the basis of. As used in this application, this term is used to describe one or more factors that affect crystals. This term does not exclude additional factors that may influence the decision. That is, the decision may be based solely on such factors, or at least in part, on such factors. Consider A based on clause "B ". Although B may be a factor affecting the decision of A, such a clause also does not preclude determining A from based on C. In other instances, A may be determined based only on B,
The scope of the present invention is not limited to any feature of the features described herein or any combination thereof (whether explicitly or implicitly), whether or not it alleviates any or all of the problems encountered in the present application, Or any generalization of them. Accordingly, the new claims may be made during the course of the present application (or a priority application) for any such combination of features. In particular, with reference to the appended claims, features in dependent claims can be combined with features in independent claims, and features in each independent claim can be combined in any suitable manner, and only in the appended claims It may not be combined with the specific combinations listed.

Various embodiments of checkpoint avoidance on the entire system for a distributed database system are disclosed. The storage node of the distributed storage system may, in some embodiments, receive one or more redo log records linked to a particular data page stored in the storage node from the database system. A data page can be one of a plurality of data pages storing data for a database. The merge event may be detected, at least in part, for a particular data page based on one or more redo log records linked to a particular data page. The merge operation may be performed to apply one or more log records to a previously stored version of a particular data page to produce a particular data page in its current state.

Various embodiments of fast failover for a distributed database system are disclosed. The database system head node may, in some embodiments, perform a failure recovery operation. When recovering from a system failure, connections with storage nodes of the distributed storage system that store data for the database may be established. In some embodiments, upon establishing connections with storage nodes, the database head node may make a database available for access. In at least some embodiments, one or more access requests may be received, and the current state of one or more data pages may be requested and received from the storage nodes.

The present specification first describes an exemplary web service based database service configured to implement checkpoint avoidance (e.g., create, delete, use, manipulate, etc.) and fast failover techniques on the entire system. The description of an exemplary web service based database service includes various aspects of an exemplary web service based database service, such as a database engine and an individual distributed database storage service. The present specification then describes flowcharts of various embodiments of methods for checkpoint avoidance and fast failover recovery on the entire system. Next, this disclosure describes an exemplary system that may implement the disclosed techniques. Various examples are provided throughout this specification.

The systems described in this application may, in some embodiments, implement a web service that enables clients (e. G., Subscribers) to operate a data storage system in a cloud computing environment. In some embodiments, the database system may be a highly variable and scalable enterprise class database system. In some embodiments, queries may be directed to a database store that is distributed across multiple physical resources, and the database system may scale up or scale down based on the need. The database system may, in different embodiments, work effectively with the database schema of the various types and / or structures. In some embodiments, clients / subscribers can submit queries in many ways, for example, bidirectionally via an SQL interface to the database system. In other embodiments, external applications and programs may submit queries using Open Database Connectivity (ODBC) and / or Java Database Connectivity (JDBC) driver interfaces to the database system.

More specifically, the systems described in this application may, in some embodiments, implement a service oriented database architecture in which the various functional components of a single database system are essentially distributed. For example, lashing together a number of complete and monolithic database instances, each of which may include an application server, a search function, or other functionality beyond what is required to provide the database's core capabilities Rather, these systems can structure the basic operations of the database (e.g., query processing, transaction management, caching and storage) into tiers that can be individually and independently variable. For example, in some embodiments, each database instance in the systems described in the present application may include a database tier (which may include a single database engine head node and a client-side storage system driver) System (which may include multiple storage nodes that collectively perform some of the operations traditionally performed in the database tier of existing systems).

(E.g., backup, recovery, snapshot, restore, log record operations, and / or various spatial management operations) of the lowest level operations of the database may be stored Lt; / RTI > and may be distributed across multiple nodes and storage devices. For example, in some embodiments, a change to a stored database (or their data pages), rather than a database engine that sends changes to the database (or their data pages) May be the responsibility of the storage layer itself. In such embodiments, redo log records, rather than modified data pages, may be sent to the storage layer, and then redo (e.g., application of redo log records) may be performed somewhat slower and in a distributed manner , By a background process). In some embodiments, failover (e.g., reconfiguration of data pages from stored redo log records) may also be performed by the storage layer and may also be performed in a distributed (and, in some cases, slower) background process ≪ / RTI >

In some embodiments, there may be significantly less network traffic between the database tier and the storage tier than existing database systems, since only redo logs (and not modified data pages) are sent to the storage tier. In some embodiments, each redo log may be approximately one-tenth the size of the corresponding data page for which it specifies changes. Note that requests sent from the database tier and distributed storage system may be asynchronous, and that multiple such requests can occur simultaneously.

Generally, after a piece of data is given, the main requirement of the database is that it can eventually return the corresponding piece of data. To this end, the database may include several different components (or tiers), each of which performs a different function. For example, a traditional database may include three tiers: a first tier for performing query analysis, optimization and execution; A second tier for providing transactionality, recovery, and persistence; And a third tier that provides storage on locally-connected disks or on a network-attached storage. As mentioned above, previous attempts to scale traditional databases typically involved replicating all three tiers of the database and distributing such replicated database instances across multiple machines.

In some embodiments, the systems described in this application may distribute the functionality of the database system differently than a traditional database, and may only provide a subset of functional components (not a complete database instance) It can be distributed across machines. For example, in some embodiments, the client-facing tier may be configured to receive a request specifying data to be stored or retrieved, rather than a method of storing or retrieving data. Such a tier may perform request analysis and / or optimization (e.g., SQL analysis and optimization) while another tier may be responsible for query execution. In some embodiments, the third tier may be responsible for providing transactionality and consistency of results. For example, such a tier may be configured to enhance the atomicity of transactional targets that target the database, while maintaining some of the so-called ACID attributes, in particular, maintaining consistency within the database and ensuring isolation between transactions targeting the database have. In some embodiments, the fourth tier may then be responsible for providing persistence of stored data in the presence of various types of failures. For example, such a tier may be responsible for change logging, recovery of database failures, managing access to primary storage volumes, and / or space management in primary storage volumes.

Referring now to the drawings, Figure 1 is a block diagram illustrating various components of a database software stack, according to one embodiment. As illustrated in this example, a database instance may comprise a number of functional components (or layers), each of which provides a portion of the functionality of the database instance. In this example, the database instance 100 includes a query analysis and query optimization layer (shown at 110), a query execution layer (shown at 120), a transactionality and consistency management layer (shown at 130) Management layer (shown as 140). As mentioned above, in some existing database systems, scaling a database instance may be performed once or several times (including all layers illustrated in FIG. 1) after a full database instance has been replicated, Which may involve adding logic. In some embodiments, the systems described in this application may instead share the functionality of persistence and space management layer 140 from a database tier to an individual storage layer, and may distribute the functionality across multiple storage nodes at the storage layer Can be dispersed.

In some embodiments, the database systems described in this application may maintain a substantial number of structures in the upper half of the database instance illustrated in FIG. 1, but may be configured to perform various operations such as backup, restore, snapshot, Responsibility for at least parts can be redistributed to the storage tier. Tightly combining redistribution of functionality in this way and log processing between database tiers and storage tiers improves performance, increases availability, and reduces costs when compared to previous approaches to providing a variable database . For example, network and input / output bandwidth requirements may be reduced because only redo log records (much smaller in size than actual data pages) may be transported across the nodes or within the delay path of the write operations. . Also, the generation of data pages may be completed independently in the background on each storage node (such as allowing for superior processing), without blocking incoming recording operations. In some embodiments, the use of a log-structured, non-overwritten repository may be achieved by using metadata manipulation, not backup, restore, snapshots, point-in-time recovery and, And may allow volume expansion operations to be performed more efficiently. In some embodiments, the storage layer may also be responsible for replicating the stored data on behalf of clients (and / or metadata associated with the data, such as redo log records) across multiple storage nodes . For example, data (and / or metadata) may be stored locally (e.g., within a single "available zone" where the collection of storage nodes runs on its own physically separate, / RTI ID = 0.0 > and / or < / RTI > across a single zone or across available zones in different zones.

In various embodiments, the database systems described in this application may support standard or on-demand application programming interfaces (APIs) for various database operations. For example, the API may include operations for creating a database, operations for creating a table, operations for altering a table, operations for creating a user, operations for dropping a user, Operations for inserting rows, operations for selecting data from within a table (e.g., operations for querying a table), operations for canceling or aborting a query, operations for creating a snapshot and / Or other operations.

In some embodiments, the database tier of a database instance receives read and / or write requests from various client programs (e.g., applications) and / or subscribers (users) And a database engine head node server that develops an execution plan to perform the database operation (s). For example, a database engine head node may develop a series of steps needed to obtain results for complex queries and joins. In some embodiments, the database engine head node is capable of managing database tiers and communication between clients / subscribers, as well as database tiers and individual distributed database-optimized storage systems.

In some embodiments, the database engine head node is responsible for receiving SQL requests from end clients over a JDBC or ODBC interface and performing locally SQL processing and transaction management (which may include locking) Can be. However, rather than locally generating data pages, the database engine head node (or its various components) can create redo log records and they can be transported to the appropriate nodes of an individual distributed storage system. In some embodiments, the client-side driver for the distributed storage system may be a storage system node (not shown) that is hosted on the database engine head node and stores segments (or data pages thereof) to which their redo log records are directed. (Or nodes) to redo log records. For example, in some embodiments, each segment may be mirrored (or otherwise sustainably) on a plurality of storage system nodes forming a protection group. In such embodiments, the client-side driver may keep track of the nodes where each segment is stored when the client request is received and may route the redo logs to all nodes where the segment is stored (e.g., asynchronously and asynchronously) In parallel, substantially simultaneously). As soon as the client-side driver receives an acknowledgment (which may indicate that a redo log record has been written to the storage node) from the write quorum of the storage nodes in the protection group, it is sent to the database tier ) To send an acknowledgment of the requested change. For example, in embodiments in which data is made sustainable through the use of protection groups, the database engine head node may continue to use the protection node until the client-side driver receives a reply from sufficient storage node instances to configure the write quorum And you may not be able to commit the transaction unless you receive it. Similarly, for a read request directed to a particular segment, the client-side driver may route the read request to all nodes where the segment is stored (e.g., asynchronously and in parallel, substantially simultaneously). As soon as the client-side driver receives the requested data from the read quorum of the storage nodes in the protection group, it may return the requested data to the database tier (e.g., to the database engine head node).

In some embodiments, a database tier (or more specifically, a database engine head node) may include a cache in which recently accessed data pages are temporarily maintained. In such embodiments, in addition to transporting the corresponding redo log record to the storage layer, when a write request is received that targets a data page held in such a cache, the Database Engine copies a copy of the data page held in its cache You can apply changes to However, unlike in database systems, a data page maintained in a cache can not be flushed to the storage layer at all, and it can be stored at any time (for example, storing a redo log record for the most recently applied write request in a cached copy At any time after being sent to the layer and acknowledged). In different embodiments, the cache may implement any of a variety of locking mechanisms to control access to the cache by as many as one writer (or multiple readers) at a time. Note, however, that in embodiments including such a cache, the cache may not be distributed across multiple nodes, but may only reside on a database engine head node for a given database instance. Thus, there may not be any cache coherency issues to manage.

In some embodiments, the database tier may support the use of synchronous or asynchronous read copies in the system, e.g., read-only copies of data on different nodes of the database tier where the read request may be routed. In such embodiments, when a database engine head node for a given database receives a read request directed to a particular data page, it routes a request for any of these read-only copies (or a particular copy) . In some embodiments, the client-side driver at the database engine head node may be configured to notify these other nodes of updates and / or invalidations to cached data pages (e.g., Then they may request updated copies of the updated data pages from the storage layer).

In some embodiments, a client-side driver operating on a database engine head node may expose a private interface to a storage tier. In some embodiments, it may also expose a traditional iSCSI interface to one or more other components (e.g., other database engines or virtual computing services components). In some embodiments, the repository for the database instance in the storage tier can be sized without limitation and modeled as a single volume that can have IOPS associated with it without limitation. When a volume is created, it is stored in a specific size, with a specific availability / persistence characteristic (e.g., specifying how it is replicated), and / or an IOPS rate associated therewith (e.g., ). ≪ / RTI > For example, in some embodiments, a variety of different persistence models may be supported, and users / subscribers may be able to determine, for their databases, how many replicated copies, zones, and / And whether it is synchronous or asynchronous based on cost objectives.

In some embodiments, the client-side driver may maintain asynchronous requests to each of the storage nodes necessary to fulfill the read requests and write requests without requiring additional hops (hops) between the storage nodes, Direct transmission is possible. For example, in some embodiments, in response to a request to change the database, the client-side driver may determine to determine one or more nodes implementing the repository for the target data page, And redirect the specified redo log record (s). The storage nodes may then be responsible for applying the specified change in the redo log record to the target data page at some future point in time. As the records are acknowledged back to the client-side driver, the client-side driver can proceed to the point where the volume is sustainable and can acknowledge commits back to the database tier. As noted previously, in some embodiments, the client-side driver may not transmit data pages at all to the storage node servers. This may not only reduce network traffic, but may also eliminate the need for checkpoint or background register threads to limit the superior-processing throughput in previous database systems.

In some embodiments, many read requests may be provided by the database engine head node cache. However, because large failure events are so common that only intra-memory replication is allowed, write requests may require persistence. Thus, the systems described in this application include a small attachment-only log-structure area in which they are recorded when two areas: redo log records are received from the database tier, and log records that create new versions of data pages in the background To minimize the cost of redo log record write operations in the dominant latency path by implementing a datastore in the storage tier as a larger area merged together for < RTI ID = 0.0 > a < / RTI > In some embodiments, an in-memory structure may be maintained for each data page pointing to the last redo log record for that page, connecting the log records backwards until the instantiated data block is referenced. This approach can be included in applications where reads are primarily cached to provide good performance for mixed read-write workloads.

In some embodiments, since the accesses to the log-structured data store for the redo log records are made up of a series of sequential input / output operations (rather than random input / output operations), the changes made are tightly packed together . Compared to existing systems where each change to a data page results in two input / output operations (one for the redo log record and one for the modified data page itself) for persistent data storage, in some embodiments, It should also be noted that the systems described in this application can prevent such "write extensions" by merging data pages into the storage nodes of the distributed storage system based on the reception of redo log records.

As noted previously, in some embodiments, the storage tier of the database system may be responsible for taking database snapshots. Taking a snapshot of a data page (e.g., a data block), however, does not result in the creation of a timestamp (or data) associated with the most recently applied redo log record in the data page / block, (E.g., a timestamp associated with the most recent operation to merge multiple redo log records to create a new version of the page / block), and recording the previous version of the page / block and any subsequent log entries And preventing garbage collection. In such embodiments, taking a database snapshot may not require reading, copying, or writing a block of data, as may be required when employing an off-volume backup strategy. In some embodiments, even though the user / subscribers may be able to choose how much additional space they want to maintain for the on-volume snapshots in addition to the active data set, only the modified data may require additional space Therefore, the space requirements for snapshots can be minimal. In different embodiments, the snapshots may be separate (e.g., each snapshot may provide access to all data in the data page from a particular point in time) or contiguous (e.g., And may provide access to all versions of data present in the inter-view data page). In some embodiments, returning to a previous snapshot may include recording the log record to indicate all the redo log records and data pages since the snapshot is invalid and available for collection, and that all database cache entries after the snapshot point And discarding. In these embodiments, it is possible to apply redo log records across the entire node in the background and to the data blocks as required, as if the storage system were performing normal forward write / read processing, No roll-forward may be required. By doing so, failover can be done and distributed across nodes in parallel.

One embodiment of a service system architecture that may be configured to implement a web services based database service is illustrated in FIG. In the illustrated embodiment, a number of clients (shown as database clients 250a through 250n) may be configured to interact with the web services platform 200 via the network 260. Web service platform 200 may be configured to interface with one or more instances of database service 210, distributed database-optimized storage service 220, and / or one or more other virtual computing services 230. It should be noted that where more than one instance of a given component may be present, reference may be made to that component in the present application, either singly or in any combination. However, the use of any one form is not intended to exclude other forms.

In various embodiments, the components illustrated in FIG. 2 may be implemented within computer hardware (e.g., by using computer hardware, such as a microprocessor or computer system), as instructions executable, directly or indirectly, As shown in FIG. For example, the components of FIG. 2 may be implemented by a system comprising a plurality of computing nodes (or simply, nodes), each of which is similar to the computer system embodiment illustrated in FIG. 10 and described below . In various embodiments, the functionality of a given service system component (e.g., a component of a database service or a component of a storage service) may be implemented by a particular node or may be distributed across several nodes. In some embodiments, a given node may implement the functionality of more than one service system component (e.g., more than one database service system component).

Generally speaking, the clients 250 are connected to the web service platform 200 via the network 260, including requests for database services (e.g., requests to create snapshots, etc.) And may include any type of client that is configurable to submit service requests. For example, a given client 250 may include a suitable version of a web browser, or may be a plug-in module or other type of code configured to run as an extension in an execution environment, Modules. Alternatively, client 250 (e.g., a database service client) may use persistent storage resources to store and / or access a database application (or its user interface), a media application, an office application, Lt; RTI ID = 0.0 > and / or < / RTI > In some embodiments, such applications may provide sufficient protocol support (e.g., Hypertext Transfer Protocol (HTTP)) to generate and process web service requests without necessarily implementing full browser support for all types of web- For the appropriate version). That is, the client 250 may be an application configured to interact directly with the web services platform 200. In some embodiments, client 250 may be configured to generate Web service requests in accordance with a Representational State Transfer (REST) -based Web service architecture, a document or message based Web service architecture, or other suitable Web services architecture.

In some embodiments, client 250 (e.g., database service clients) may be configured to provide access to web services based repositories of databases to such applications in a manner that is transparent to other applications. For example, client 250 may be configured to integrate with an operating system or file system to provide storage in accordance with a suitable variant of the storage models described in this application. However, an operating system or file system may provide a different storage interface to applications, such as the normal file system hierarchy of files, directories and / or folders. In such an embodiment, applications may not need to be modified to use the storage system service model of FIG. Instead, details of interfacing to the web services platform 200 may be adjusted by the operating system or file system and client 250 on behalf of applications executing within the operating system environment.

Clients 250 communicate web service requests (e.g., a snapshot request, a request for parameters of a snapshot, a read request, a snapshot recovery, etc.) to the web service platform 200 via the network 260 And receive responses therefrom. Network 260 may include any suitable combination of networking hardware and protocols required to establish web-based communications between clients 250 and platform 200. In some embodiments, For example, the network 260 may generally include service providers that collectively implement various communication networks and the Internet. Network 260 may also include public or private wireless networks as well as private networks such as local area networks (LANs) or wide area networks (WANs). For example, a given client 250 and web service platform 200 may each be provided within an enterprise having their own internal networks. In such an embodiment, the network 260 may include hardware (e.g., modems, routers, switches, etc.) required to establish a network 250 between a given client 250 and the Internet as well as between the Internet and the web services platform 200 , Load balancers, proxy servers, etc.) and software (e.g., protocol stacks, accounting software, fire / security software, etc.). In some embodiments, the clients 250 may communicate with the web services platform 200 using a private network rather than the public Internet. For example, clients 250 may be provided within the same enterprise as a database service system (e.g., a system that implements database service 210 and / or distributed database-optimized storage service 220). In such a case, the clients 250 can communicate with the platform 200 solely via the private network 260 (e.g., a LAN or WAN where the public is not accessible but can use Internet-based communication protocols).

Generally speaking, the web service platform 200 can be configured to implement one or more service endpoints configured to receive and process web service requests, such as requests for access to data pages (or their records) have. For example, the web services platform 200 may comprise hardware and / or software configured to implement a particular endpoint so that the HTTP-based web service request directed to that endpoint is properly received and processed. In one embodiment, web service platform 200 receives web service requests from clients 250 and provides them to database service 210, distributed database-optimized storage service 220 and / or another computing Or a server system configured to transmit to components of a system that implements the service 230. In other embodiments, the web services platform 200 may be implemented in a number of separate systems (e.g., in a cluster topology) that implement load balancing and other request management ≪ / RTI > features. In various embodiments, the web services platform 200 may be configured to support web services requests of the REST or document based (e.g., SOAP based) types.

In addition to serving as addressable endpoints for client web service requests, in some embodiments, web service platform 200 may implement various client management features. For example, the platform 200 may determine the identity of the requesting clients 250, the number and / or frequency of client requests, the database tables (or their records) stored or retrieved on behalf of the clients 250, By tracking the total storage bandwidth used by clients 250, the class of storage requested by clients 250, or other measurable client usage parameters, Weighing and calculation of usage can be adjusted. The platform 200 may also implement financial accounting and billing systems or may maintain a database of usage data that may be queried and processed by external systems for reporting and accounting for client usage activity. In some embodiments, the platform 200 may include an indicator that reflects the types and rates of requests received from the clients 250, the bandwidth utilized by those requests, the system processing latency for such requests, (E.g., network bandwidth and / or storage utilization within the storage service system), rates and types of errors resulting from requests, properties of the stored and requested data pages or their records Monitoring, and / or aggregating various storage service system operational indicators, such as, for example, the size, type, data type, etc., or any other suitable indicator. In some embodiments, such an indicator may be used by system administrators to coordinate and maintain system components, while in other embodiments such an indicator (or related portions of such an indicator) Exposure to clients to monitor their usage for database services 210, distributed database-optimized storage services 220 and / or other virtual computing services 230 (or the underlying systems that implement such services) .

In some embodiments, the platform 200 may also implement user authentication and access control procedures. For example, for a given web service request for access to a particular database, the platform 200 may be configured to verify that the client 250 associated with the request is authorized to access a particular database. The platform 200 may be configured to evaluate the requested access to a particular database, for example by evaluating an identity, password or other qualification against the qualifications associated with a particular database, or against an access control list for a particular database, Authentication can be determined. For example, if the client 250 does not have sufficient credentials to access a particular database, the platform 200 may return a response indicating the error status to the requesting client 250, for example, You can refuse. The various access control policies may be stored in records or lists of access control information by the database service 210, the distributed database-optimized storage service 220, and / or other virtual computing services 230.

The web services platform 200 may be configured to allow the client 250 to represent the primary interface that can access the features of the database system that implements the database service 210 but that it does not need to represent a unique interface to such features . For example, an alternative API, which may be separate from the web service interface, may be used to allow clients inside the enterprise providing the database system to bypass the web service platform 200. In many of the examples described in this application, the distributed database-optimized storage service 220 may reside within a computing system or enterprise system that provides database services to clients 250, and may include external clients (e.g., Or client applications). In such embodiments, an internal "client" (e.g., database service 210) is coupled to a distributed database-optimized storage service 220 and a database service 210 via a local or private network, (E.g., via APIs, directly between systems implementing these systems) to access the distributed database-optimized storage service 220. In such embodiments, the use of the distributed database-optimized storage service 220 in storing the database on behalf of the clients 250 may be transparent to such clients. In other embodiments, the distributed database-optimized storage service 220 may be used by the web service platform 200 to provide a repository of databases or other information for applications that are not dependent on the database service 210 for database management. Lt; RTI ID = 0.0 > 250 < / RTI > This is illustrated in FIG. 2 by the dotted line between the web service platform 200 and the distributed database-optimized storage service 220. In these embodiments, clients of the distributed database-optimized storage service 220 may access the distributed database-optimized storage service 220 via the network 260 (e.g., via the Internet). In some embodiments, the virtual computing service 230 may receive from the distributed database-optimized storage service 220 (e. G., For example, a < / RTI > , Virtual computing service 230, and distributed database-optimized storage service 220). This is illustrated in FIG. 2 by the dotted line between the virtual computing service 230 and the distributed database-optimized storage service 220. In some cases, the account and / or credential services of the platform 200 may be unnecessary for intra-service components within the same enterprise or for internal clients such as administrative clients.

In various embodiments, different storage policies may be implemented by the database service 210 and / or the distributed database-optimized storage service 220. Examples of such storage policies include persistence policies (e.g., policies representing multiple instances of the database (or data pages thereof) to be stored and a number of different nodes where they will be stored) and / or load balancing policies Which may distribute databases, or data pages thereof, across different nodes, volumes and / or disks in an attempt to equalize). In addition, different storage policies may be applied to different types of stored items by one of a variety of services. For example, in some embodiments, the distributed database-optimized storage service 220 may implement a higher persistence for redo log records than data pages.

3 is a block diagram illustrating various components of a database system including a database engine and an individual distributed database storage service, in accordance with one embodiment. In this example, the database system 300 includes a plurality of database engine head nodes 320 and a distributed database-optimized storage service 310 (each of which is shown as database clients 350a through 350n) Which may or may not be visible to clients). As illustrated in this example, one or more database clients 350a-350n may be network-addressable and accessible to the database clients 350a-350n via network 360 (e.g., (E.g., head node 320a, head node 320b, or head node 320c, each of which is a component of each database instance) . However, to store data pages (and redo log records and / or other metadata associated with them) of one or more databases on behalf of the database clients 350a-350n, The distributed database-optimized storage service 310, which may be employed by the database system to perform other functions, may, in different embodiments, be network-addressable and accessible to the storage clients 350a-350n It may or may not be possible. For example, in some embodiments, the distributed database-optimized storage service 310 may provide various storage, access, change logging, recovery, log record manipulation, and / or space saving operations to the storage clients 350a-350n in an invisible manner. Management operations.

As previously mentioned, each database instance receives requests (e.g., a snapshot request) from various client programs (e.g., applications) and / or subscribers (users) A single database engine head node 320 that analyzes and optimizes them and develops an execution plan to perform the associated database operation (s). In the example illustrated in FIG. 3, the query analysis, optimization, and execution component 305 of the database engine head node 320a receives a query from the database client 350a and the database engine head node 320a is a component database instance Lt; RTI ID = 0.0 > queries. ≪ / RTI > In some embodiments, the query analysis, optimization, and execution component 305 may return query responses to the database client 350a, which may suitably include record acknowledgments, requested data pages (Or portions thereof), error messages, and / or other responses. As illustrated in this example, the database engine head node 320a may also include a client-side storage service driver 325, which is responsible for reading requests to various storage nodes in the distributed database- Optimize storage services 310, receive requested data pages from the distributed database-optimized storage service 310, and / or receive data pages from the distributed database- Pages, error messages, or other responses to the query analysis, optimization, and execution component 305 (which may, as a result, return them to the database client 350a).

In this example, the database engine head node 320a includes a data page cache 335, where recently accessed data pages may be temporarily held. As illustrated in Figure 3, the database engine head node 320a may also include a transaction and consistency management component 330, which may be a database engine head node 320a, And to provide consistency. For example, these components may be responsible for ensuring the atomicity, consistency, and isolation properties of the database instance and the transactions indicated to that database instance. As illustrated in Figure 3, the database engine head node 320a may also include a transaction log 340 and an undo log 345, which tracks the status of various transactions, May be employed by the transaction and consistency management component 330 to roll back any locally cached results of the transaction.

Each of the other database engine head nodes 320 illustrated in FIG. 3 (e.g., 320b and 320c) may include similar components and may be received by one or more database clients 350a-350n, Note that you can perform similar functions for queries directed to each database instance.

In some embodiments, the distributed database-optimized storage systems described in this application may structure data into various logical volumes, segments, and pages for storage on one or more storage nodes. For example, in some embodiments, each database may be presented by a logical volume, and each logical volume is segmented through a collection of storage nodes. Each segment of the storage nodes, which resides on a particular storage node, contains a set of contiguous block addresses. In some embodiments, each data page is stored in a segment such that each segment contains a collection of one or more data pages and a change log (also referred to as a redo log) for each data page it stores (e.g., Records of records). As described in detail in this application, the storage nodes may be configured to receive redo log records (which may also be referred to as ULRs in this application) and to store new versions of corresponding data pages and / (E. G., In response to a slow and / or request for a data page or a database failure) to merge redo log records to create the log records. In some embodiments, data pages and / or change logs may be mirrored across multiple storage nodes, depending on various configurations (which may be specified by databases maintained on the database system on behalf of the client) . For example, in different embodiments, one, two, or three copies of data or change logs may have one, two, or three different, depending on the default configuration, application specific persistence capabilities, or client explicit persistence capabilities And may be stored in each of the available zones or areas.

As used in this application, the following terms may be used to describe the structuring of data by a distributed database-optimized storage system, according to various embodiments.

Volume: A volume is a logical concept that represents a very persistent unit of storage that is understood by the user / client / application of the storage system. More specifically, a volume is a distributed store that appears in a user / client / application as a single, consistent sequence of log operations of the write operations to the various user pages of the database table. Each recording operation can be encoded with a user log record (ULR), which represents a logical, ordered variation on the content of a single user page in the volume. As mentioned above, the ULR may also be referred to as a redo log record in the present application. Each URL may include a unique identifier (e.g., a Logical Sequence Number (LSN)) that it may include. Each URL may include a protection group (PG) to provide high persistence and availability to the URL. Protection Group). The volume may provide bytes of a variable-size contiguous range to an LSN-type read / write interface.

In some embodiments, the volumes may be configured in multiple ranges, each sustainable through a protection group. In these embodiments, the volume may represent a unit of storage comprised of volume ranges of two adjacent sequences. Reads and writes pointed to the volume can be mapped to readings and writes corresponding to the constituent volume ranges. In some embodiments, the size of the volume can be changed by adding or removing volume ranges from the end of the volume.

Segment: A segment is a restricted persistence unit of a repository that is assigned to a single storage node. More specifically, a segment provides limited best effort persistence (e.g., a permanent, non-redundant single point of failure that is a storage node) for a particular fixed size byte of data. Such data may in some cases be a mirror of user addressable data, or it may be other data, such as volume metadata or erase coded bits, in various embodiments. The predetermined segment may exist in exactly one storage node. Within the storage node, multiple segments may reside in each SSD, and each segment may be limited to one SSD (e.g., a segment may not extend across multiple SSDs). In some embodiments, the segment may not be required to occupy a contiguous area on the SSD; Rather, there may be an allocation map in each SSD that describes the area owned by each of the segments. As mentioned above, the protection group may be composed of a plurality of segments spread over a plurality of storage nodes. In some embodiments, the segment may provide a fixed-size contiguous range of bytes to an LSN-type read / write interface where the size is defined at creation. In some embodiments, each segment may be identified by a segment UUID (e.g., a universally unique identifier of the segment).

Storage page: The storage page is a block of memory, which usually has a fixed size. In some embodiments, each page is a block (e.g., of virtual memory, disk, or other physical memory) of a size of memory defined by the operating system and is also referred to in the present application by the term & . More specifically, the store page may be a set of contiguous sectors. It can be an allocation unit in SSDs, as well as a unit in log pages where headers and metadata are present. In some embodiments and in the context of database systems described in this application, the term "page" or "storage page" may refer to a similar block of the size defined by the database configuration, typically 4096, 8192, 16384, or 32768 bytes.

Log page: The log page is a type of storage page used for storage log records (for example, redo log records or undo log records). In some embodiments, the log pages may be the same size as the storage pages. Each log page may include a header that includes meta data for that log page, e.g., metadata identifying the segment to which it belongs. Note that the log page is a structured unit and may not be the data unit involved in the write operations. For example, in some embodiments, during normal forward processing, write operations may write one sector to the tail of the log at the same time.

Log records: Log records (e.g., individual elements of a log page) may have several different classes. For example, user log records (URLs), which may be generated and understood by users / clients / applications of the storage system, may be used to represent changes to user data in the volume. The control log records (CLRs) generated by the storage system may include control information used to keep track of metadata, such as the current unconditional volume sustainable LSN (VDL). Null log records (NLRs) may be used as padding to fill unused space in log sectors or log pages in some embodiments. In some embodiments, there may be log records of various types within each of these classes, and the type of log record may correspond to the function that should be applied to interpret the log record. For example, one type may represent all data of a user page in a compressed format using a particular compression format; The second type may represent new values of byte range within the user page; The third type may represent an incremental operation on a sequence of bytes that are interpreted as integers; The fourth type may include copying one byte range to another location within the page. In some embodiments, log record types may be identified by GUIDs (not by integers or enums), which may simply be versioning and development, especially for ULRs .

Payload: The payload of a log record is data or parameter values that are specific to a particular type of log record or log record. For example, in some embodiments, there may be a majority of (or all) log records, and a set of parameters or attributes that the storage system itself understands. These attributes can be part of the common log record header / structure, which can be relatively small when compared to the sector size. In addition, most log records may contain additional parameters or data specific to that log record type, and such additional data may be regarded as a payload of that log record. In some embodiments, if the payload for a particular URL is greater than the user page size, it may be replaced by an absolute URL (AURL) whose payload contains all the data for the user page. This may enable the storage system to enhance the upper bound on the size of the payload for the same ULRs as the size of the user pages.

Note that when the stored log records are in the segment log, the payload may, in some embodiments, be stored with the log header. In other embodiments, the payload may be stored in a separate location, and pointers to where the payload is stored may be stored with the log header. In still other embodiments, a portion of the payload may be stored in the header, and the remainder of the payload may be stored in a separate location. If the entire payload is stored with a log header, this can be referred to as in-band storage; Other stores may be referred to as out-of-band. In some embodiments, the payloads of most large AULRs may be stored out of band at the cold zone of the log (to be described below).

User pages: User pages are byte ranges (of fixed size) and their sorts for a particular volume visible to users / clients of the storage system. User pages are logical concepts, and the bytes of certain user pages may or may not be stored in any storage page. The size of the user pages for a particular volume may be independent of the storage page size for that volume. In some embodiments, the user page size may be configurable per volume, and different segments on the storage node may have different user page sizes. In some embodiments, user page sizes may be limited to be a multiple of a sector size (e.g., 4 KB) and may have an upper limit (e.g., 64 KB). On the other hand, the stored page size may be fixed for the entire storage node and may not change as long as there is no change to the underlying hardware.

Data page: A data page is a type of storage page used to store user page data in a compressed format. In some embodiments, every piece of data stored in a data page is associated with a log record, and each log record may include a pointer to a sector within the data page (also referred to as the data sector). In some embodiments, the data pages may not contain any built-in metadata that is not provided by each sector. There may be no relationship between the sectors in the data page. Instead, the structuring into pages may simply exist as a representation of the segmentation of the allocation of data to the segment.

Storage node: The storage node is a single virtual machine in which the storage node server code is located. Each storage node may comprise a number of locally connected SSDs and may provide a network API for access to one or more segments. In some embodiments, the various nodes may be on the active list or on a degraded list (e.g., if they are slow or otherwise damaged in response, but not completely unusable). In some embodiments, the client-side driver is used to determine when nodes should be replaced and when to be replaced, and / or when redistributing data among the various nodes based on observed performance and / (Or may be) responsible for classifying the nodes as active or degraded to determine how to do so.

SSD: As used herein, the term "SSD" refers to a storage volume, such as a disk, solid state drive, battery backed RAM, non-volatile RAM device (e.g., one or more NV- , Or may represent a local block storage volume as seen by the storage node, regardless of the type of storage employed by another type of persistent storage device. SSDs do not necessarily have to be directly mapped to hardware. For example, a single solid state storage device may be distributed and / or distributed across multiple local volumes where each volume is divided into multiple segments and spanned across multiple segments, or a single drive may simply be easier to manage Lt; RTI ID = 0.0 > volume. ≪ / RTI > In some embodiments, each SSD may store an allocation map in a single fixed location. Such a map may indicate which storage pages are owned by particular segments, and which of these are log pages (not data pages). In some embodiments, storage pages may be preallocated to each segment such that forward processing may not need to wait for allocation. Any changes to the allocation map may have to be made sustainable before the newly allocated storage pages are used by the segments.

One embodiment of a distributed database-optimized storage system is illustrated by the block diagram of FIG. In this example, database system 400 includes a distributed database-optimized storage system 410, which communicates with database engine head node 420 via interconnect 460. As in the example illustrated in FIG. 3, the database engine head node 420 may include a client-side storage service driver 425. In this example, the distributed database-optimized storage system 410 includes a plurality of storage system server nodes 430 (including those shown as 440, and 450), each of which is a storage for the segment Storage for data pages and redo logs, and hardware and / or software configured to perform various management functions. For example, each storage system server node may perform the following operations: replication (locally, e.g. in a storage node), merging of redo logs to create data pages, snapshots At least a portion of any or all of the operations (e.g., recovery, deletion, etc.), log management (e.g., manipulating log records), failover, and / And / or software configured to perform the functions described herein. Each storage system server node may also include a number of connection storage devices (e.g., SSDs) that data blocks may be stored on behalf of clients (e.g., users, client applications, and / or database service subscribers) Lt; / RTI >

4, storage system server node 430 includes data page (s) 433, segment redo log (s) 435, segment management functions 437, and connection SSDs 471, 478). Note again that the label "SSD" may or may not represent a solid state drive, but may more generally represent a local block storage volume, regardless of its underlying hardware. Similarly, storage system server node 440 includes data page (s) 443, segment redo log (s) 445, segment management functions 447, and connection SSDs 481 through 488 do; The storage system server node 450 includes data page (s) 453, segment redo log (s) 455, segment management functions 457, and connection SSDs 491-498.

As noted previously, in some embodiments, a sector may be the largest unit on the SSD, which is an alignment unit on the SSD and the write can be recorded only partially without risk of being completely complete. For example, the sector size for various solid state drives and spinning media may be 4 KB. In some embodiments of the distributed database-optimized storage systems described in this application, each and every sector may have a 64-bit (8 byte) CRC at the beginning of the sector, regardless of the higher level entity that the sector is a part of . In such embodiments, such a CRC (which can be authenticated every time the sector is read from the SSD) can be used to detect corruptions. In some embodiments, each and every sector may also include a "sector type" byte that identifies whether the value is a log sector, a data sector, or a sector as an uninitialized sector. For example, in some embodiments, a sector type byte value of zero may indicate that a sector has not been initialized.

In some embodiments, each of the storage system server nodes in the distributed database-optimized storage system manages communication with the database engine head node, for example, to receive redo logs, resend data pages, Lt; RTI ID = 0.0 > operating < / RTI > In some embodiments, all data blocks written to the distributed database-optimized storage system may be backed up to long term and / or archival storage (e.g., to a remote key-value sustainable backup storage system).

5 is a block diagram illustrating use of an individual distributed database-optimized storage system in a database system, in accordance with one embodiment. In this example, one or more client processes 510 may store data in one or more databases maintained by a database system that includes a database engine 520 and a distributed database-optimized storage system 530. In the example illustrated in Figure 5, the database engine 520 includes database tier components 560 and client-side driver 540 (between the distributed database-optimized storage system 530 and the database tier components 560) Which serves as an interface). In some embodiments, the database tier components 560 perform functions such as those performed by the query analysis, optimization and execution component 305 of FIG. 3 and the transaction and consistency management component 330 (Such as those stored by the data page cache 335 of Figure 3, the transaction log 340 and the undo log 345) and / or data pages, transaction logs and / or undo logs .

In this example, the one or more client processes 510 include database query requests 515 (read and / or write that target data stored on one or more storage nodes 535a through 535n) to database tier components 560 Requests) from the database tier components 560 and may send database query responses 517 (e.g., responses that include record acknowledgments and / or requested data) Lt; / RTI > Each database query request 515, including a request to write to a data page, can be analyzed and optimized to generate one or more record record requests 541, which is followed by a distributed database-optimized storage system 530 May be sent to client-side driver 540 for routing. In this example, the client-side driver 540 may generate one or more redo log records 531 corresponding to the record record request 541 and send them to the storage nodes of the distributed database-optimized storage system 530 To the particular storage nodes among the storage nodes 535. The distributed database-optimized storage system 530 may return to the database engine 520 (specifically to the client-side driver 540) a corresponding record acknowledgment 523 for each redo log record 531 . The client-side driver 540 may communicate these record acknowledgments (such as write responses 542) to the database tier components 560, which may then send the database query responses 517 as one or more (E. G., Record acknowledgments) corresponding to the client process 510. < / RTI >

In this example, each database query request 515, which includes a request to read a data page, can be analyzed and optimized to generate one or more read record requests 543, which may be distributed database-optimized storage system 530 To the client-side driver 540 for subsequent routing to the client-side driver 540. [ In this example, the client-side driver 540 may send these requests to specific ones of the storage nodes 535 of the distributed database-optimized storage system 530 and the distributed database-optimized storage system 530 May return requested data pages 533 to database engine 520 (specifically to client-side driver 540). The client-side driver 540 may send the requested data pages as return data records 544 to the database tier components 560 and then the database tier components 560 may send the database query requests (517) to one or more client processes (510).

In some embodiments, various error and / or data loss messages 534 may be sent from the distributed database-optimized storage system 530 to the database engine 520 (specifically to the client-side driver 540) have. These messages may be sent to the database tier components 560 from the client-side driver 540 as error and / or loss report messages 545 and then to (or in place of) the database query response 517 To the client process 510 as described above.

In some embodiments, the APIs 531-534 of the distributed database-optimized storage system 530 and the APIs 541-545 of the client-side driver 540 allow the database engine 520 to perform distributed database- May expose the functionality of the distributed database-optimized storage system 530 to the database engine 520 as if it were a client of the storage system 530. For example, the database engine 520 (via the client-side driver 540) may perform various operations of the database system (e. G., By way of example) implemented by a combination of the database engine 520 and the distributed database-optimized storage system 530 (Or to facilitate their execution) to perform redo log records or request data pages via these APIs to perform (e.g., store, access, change logging, recovery, and / . As illustrated in FIG. 5, the distributed database-optimized storage system 530 may store blocks of data on storage nodes 535a through 535n, each of which may have multiple connected SSDs. In some embodiments, the distributed database-optimized storage system 530 may provide high persistence of stored data blocks through applications of various types of redundancy techniques.

In various embodiments, API calls and responses (e.g., APIs 531-534) between the database engine 520 and the distributed database-optimized storage system 530 of FIG. 5 and / or the client- (E.g., APIs 541-545) between database tier components 540 and database tier components 560 can be used to provide secure proxy connections (e.g., those managed by a gateway control plane) Or through a public network, or, alternatively, via a private channel, such as a virtual private network (VPN) connection. These and other APIs for and / or between the components of the database systems described in this application include, but are not limited to, Simple Object Access Protocol (SOAP) technology and Representational state transfer (REST) May be implemented according to different techniques. For example, these APIs may, but need not, be implemented as SOAP APIs or RESTful APIs. SOAP is a protocol for exchanging information in the context of Web-based services. REST is an architectural type for distributed hypermedia systems. The RESTful API (also referred to as a RESTful Web service) is a Web services API implemented using HTTP and REST technologies. The APIs described in this application may be used in some embodiments to support integration with the database engine 520 and / or distributed database-optimized storage system 530, including but not limited to C, C ++, Java, It can be wrapped in client libraries in various languages, including C # and Perl.

As mentioned above, in some embodiments, the functional components of the database system are distributed among the components performed by the database engine and the components performed in separate, distributed, database-optimized storage systems . In one particular example, in response to receiving a request from a client process (or its thread) to insert something into the database (e.g., to update a single data block by adding a record to the data block) , One or more components of the database engine head node may perform query analysis, optimization, and execution and may transmit each portion of the query to the transaction and consistency management component. The transaction and consistency management component can ensure that no other client process (or its thread) is attempting to transform the same row at the same time. For example, transaction and consistency management components may be responsible for ensuring that these changes are performed atomically, consistently, persistently, and in a manner that is separate from the database. For example, the transaction and consistency management component may be configured to generate redo log records to be sent to one of the nodes in a distributed database-optimized storage service, and in order for them and / or ACID attributes to be satisfied for such a transaction Side storage service driver of the database engine head node (with other redo log records generated in response to other client requests) for transmission to the distributed database-optimized storage service at a guaranteed timing. Upon receiving a redo log record (which may be regarded as a "update record" by the storage service), the corresponding storage node may update the data block and store a redo log record (e.g., A record of all changes being indicated). In some embodiments, the database engine may be responsible for generating the undo log record for this change, and may also include a redo log record that can be used locally (within the database tier) It may be responsible for generating both records. However, unlike traditional database systems, the systems described in this application save the responsibility for applying changes to data blocks (rather than applying the data blocks to the database and transporting the changed data blocks to the storage system) It can be transferred to the system. 8-9B, in various embodiments, a checkpoint on the entire system may also be used to determine whether or not a checkpoint from the database system failure due to various log record operations may be performed by the storage system. Can be avoided in database systems with fast recovery.

A variety of different allocation models may be implemented for the SSD in different embodiments. For example, in some embodiments, log entry pages and application pages may be allocated from a single heap of pages associated with the SSD device. This approach may have the advantage of allowing relative storage to be consumed by log pages and data pages to keep it indefinitely and to adapt automatically to usage. It may also have the advantage of allowing the pages to be kept unprepared until they are used, and to be freely adapted for other purposes without preparation. In other embodiments, the allocation model may distribute the storage device into separate spaces for log entries and data pages. Such an allocation model is illustrated by the block diagram of FIG. 6 and described below.

6 is a block diagram illustrating how data and metadata may be stored in a given storage node (or persistent storage device) of a distributed database-optimized storage system, in accordance with one embodiment. In this example, the SSD storage space 600 stores the SSD header and other fixed metadata in a portion 610 of the labeled space. It stores log pages in a portion 620 of the labeled space and includes labeled space 630 that is initialized and reserved for additional log pages. A portion of SSD storage space 600 (shown as 640) is initialized, but unassigned, and other portions of space (shown as 650) are uninitialized and unassigned. Finally, a portion 660 of the labeled SSD storage space 600 stores data pages.

In this example, the first available log page slot is referred to as 615, and the last used log page slot (short term) is referred to as 625. The last reserved log page slot is referred to as 635, and the last available log page slot is referred to as 645. In this example, the first used data page slot (short term) is referred to as 665. In some embodiments, the locations of each of these elements (615, 625, 635, 645, and 665) in the SSD storage space 600 may be identified by respective pointers.

In the allocation approach illustrated in FIG. 6, valid log pages may be packed to the beginning of the flat storage space. Halls made available due to freed log pages can be reused before additional log page slots that are further to the address space are used. For example, in the worst case, the first n log page slots contain valid log data, where n is the largest number of valid log pages that have never existed simultaneously. In this example, valid data pages may be packed at the end of the flat storage space. Halls made available due to free data pages can be reused before additional lower log page slots are used in the address space. For example, in the worst case, the first m data pages contain valid data, where m is the largest number of valid data pages that have never existed simultaneously.

In some embodiments, before a log page slot can become part of a potential set of valid log page entries, it may have to be initialized to a value that can not be confused with a valid future log entry page. This is implicitly true for recycled log page slots because the retrieved log page has enough metadata to never be confused with the new valid log page. However, when the storage device is first initialized, or when the space potentially used to store application data pages is recycled, the log page slots may have to be initialized before they are added to the log page slot pool. In some embodiments, the rebalance / recycle log space may be performed as a background task.

In the example illustrated in FIG. 6, the current log page slot pool includes an area between the first available log page slot (at 615) and the last allocated log page slot 625. In some embodiments, this pool may be safely extended to the last available log page slot 625 without re-initializing the new log page slots (e.g., the last allocated log page slots 635 By continuing to update the pointer to identify it). In this example, past the last available log page slot (identified by pointer 645), the pool continues the initialized log page slots and continues to provide pointers to the last available log page slot 645 (As identified by the pointer 665) that was originally used by the update. In this example, previously uninitialized and unallocated portions of the SSD storage space 600, shown as 650, may be directed to a service for storing log pages. In some embodiments, the current log page slot pool is moved down the location of the last used log page slot (identified by the pointer) by continuing to update the pointer 635 for the last allocated log page slot Can be reduced.

In the example illustrated in FIG. 6, the current data page slot pool includes a region between the last available log page slot (identified by pointer 645) and the end of SSD storage space 600. In some embodiments, the data page pool can be safely secured to the location identified by the pointer to the last allocated reserved page page slot 635 by continuing to update the pointer to the last available log page slot 645 Can be enlarged. In this example, the previously unassigned portion of the SSD storage space 600, shown at 640, may be directed to a service for storing data pages. The pool effectively reallocates portions of the SSD storage space 600, shown as 630 and 640, for storing data pages, rather than log pages, so that the last allocated log page slot 635 and the last used Can be safely magnified to the location identified by the last used log page slot 625 by continuing updates to pointers to possible log page slots 645. [ In some embodiments, the data page slot pool may be a pointer to the first used data page slot 665 by initializing additional log page slots and continuing to update the pointer to the last available log page slot 645 As shown in FIG.

In embodiments employing the allocation approach illustrated in FIG. 6, page sizes for log page pools and data page pools can still be selected independently, while still facilitating good packing operations. In these embodiments, there is no possibility of a valid log page linking to a spoofed log page formed by the additional data, and there is no possibility of a corrupted log and a valid log tail linking to the next page not yet written It may be possible to distinguish between the two. In embodiments employing the allocation approach illustrated in FIG. 6, at startup, all log page slots up to the location identified by the pointer to the last allocated reserved log page slot 635 are read out quickly and sequentially And the entire log index can be reconstructed (including inferred connection / sequencing). In these embodiments, there may not be any need for explicit connectivity between log pages, since everything can be deduced from the LSN sequencing constraints.

In some embodiments, the segment may consist of three main parts (or zones): one containing a hot log, one containing a cold log, and one containing user page data. The zones do not necessarily have to be contiguous regions of the SSD. Rather, they can be interspersed with subdivision of the storage page. In addition, there may be a root page for each segment that stores metadata for the segment and its attributes. For example, the root page for a segment may include the user page size for the segment, the number of user pages in the segment, the current entry / head of the hot log zone (which may be recorded in the form of a flush number), volume epoch ), And / or access control metadata.

In some embodiments, hot-log zones may receive new records from a client as they are received by the storage node. Delta user log records (DULRs) that specify changes to user / data pages in delta form from previous versions of the page, and absolute user / data page content, Both Absolute User Log Records (AULRs) can be completely recorded in the log. Log records may be added to this zone in the order in which they are received (e.g., they are not sorted by the LSN), and they may go over the log pages. Log records can be self-describing, for example, they may contain an indication of their size. In some embodiments, no garbage collection is performed in this zone. Instead, the space can be recycled by truncating all requested log records from the start of the log after being copied to the cold log. The log sectors in the hot zone may be annotated with the most recently known unconditional VDL at each time the sector is recorded. Conditional VDL CLRs may be recorded in the hot zone as they are received, but the most recently recorded VDL CLR may be significant.

In some embodiments, each time a new log page is written, it may be assigned a flush number. The flush number may be recorded as part of every sector in each log page. The flush numbers can be used to determine which log pages were later written when comparing two log pages. The flush numbers increase monotonically and are scoped to the SSD (or storage node). For example, a monotonically increasing set of flush numbers is shared among all segments on the SSD (or all segments on the storage node).

In some embodiments, in the cold log zone, log records may be stored in ascending order of their LSNs. In this zone, AULRs may not need to store data in-line, depending on their size. For example, if they have large payloads, all or a portion of the payloads may be stored in the data zone and they may indicate where their data is stored in the data zone. In some embodiments, the log pages in the cold log zone may be recorded at one time and not a sector-by-sector, complete page. Since the log pages in the cold zone are simultaneously written in complete pages, any log pages in the cold zone where the flush numbers in all sectors are not the same can be considered incompletely written pages and can be ignored. In some embodiments, in the cold log zone, DULRs may be traversed across log pages (up to two log pages). However, AULRs may not lead across log sectors, for example, so a merge operation would be able to replace the DULR with a single atomic record with an AULR.

In some embodiments, the cold log zone is populated by copying the log records from the hot log zone. In such embodiments, log records less than or equal to the current unconditional volume sustainable LSN (VDL) may be eligible to be copied only to the cold log zone. When moving log records from the hot log zone to the cold log zone, some log records (such as many CLRs) may not need to be copied because they are no longer needed. In addition, some additional merge of user pages may be performed at this point, which may reduce the amount of radiation required. In some embodiments, when a given hot zone log page has been fully written and is no longer the latest hot zone log page, and all ULRs on the hot zone log page have been successfully copied to the cold log zone, Can be freely reused.

In some embodiments, garbage collection can be performed in the cold log zone to reclaim the space occupied by the log records, e.g., log records that no longer need to be stored in the SSDs of the storage tier. have. For example, a log record may become obsolete when there is a subsequent AULR for the same user page, and the version of the user page presented by that log record is not needed for retention on the SSD. In some embodiments, the garbage collection process merges two or more contiguous log pages and replaces them with fewer new log pages containing all the obsolete log records from the log pages they are replacing It can be recycled. New log pages may be assigned new flush numbers that are larger than the flush numbers of the log pages they are replacing. After the recording of these new log pages is completed, the replaced log pages may be added to the free page pool. Note that in some embodiments, there may not be any explicit connection of log pages using arbitrary pointers. Instead, the sequence of log pages may be implicitly determined by the flush numbers on their pages. Whenever multiple copies of a log record are found, the log records present in the log page with the highest flush number may be considered valid and others may be considered obsolete.

In some embodiments, there may be some fragmentation, for example, because the granularity of the space managed in the data zone (sector) may differ from the granularity of the data zone (storage page). In some embodiments, to maintain this fragmentation being controlled, the system can keep track of the number of sectors used by each data page, preferentially allocate from data pages that have fewer free spaces, It may preferentially garbage collect almost empty data pages (which may need to move data to a new location if it is still relevant). The pages assigned to the segments may be tailored to different purposes among the three zones in some embodiments. For example, when a page that was allocated to a segment becomes free, it may remain associated with that segment for a period of time and subsequently used in any of the three zones of that segment. The sector header of all sectors may indicate the zone to which the sector belongs. When all sectors on a page are free, the page can be returned to a common free storage page pool that is shared across zones. This free store page sharing may reduce (or prevent) fragmentation in some embodiments.

In some embodiments, the distributed database-optimized storage systems described in this application can maintain various data structures in memory. For example, for each user page present in the segment, the user page table may include a bit indicating whether this user page is "removed" (i.e., whether it contains all zeros) An LSN of the latest log record, and an array / list of locations of all log records from the hot log zone for the page. For each log record, the user page table contains the sector number, the offset of the log record in the sector, the number of sectors to read in the log page, the sector number of the second log page (if the log record follows across the log pages ), The number of sectors to be read in the log page can be stored. In some embodiments, the user page table may also store an LSN of all log records from the cold log zone and / or an array of sector numbers for the payload of the latest AULR if it is in the cold log zone.

In some embodiments of the distributed database-optimized storage systems described in this application, the LSN index may be stored in memory. The LSN index can map LSNs to log pages in the cold log zone. When the corresponding log records in the cold log zone are sorted, it may be that they contain one entry per log page. However, in some embodiments, all obsolete LSNs are stored in the index and can be mapped to corresponding sector numbers, offsets, and numbers of sectors for each log record.

In some embodiments of the distributed database-optimized storage systems described in this application, the log page table may be stored in memory, and the log page table may be used during garbage collection of the cold log zone. For example, the log page table can identify which log records are useless (for example, which log records can be garbage collected) and how free space is available on each log page.

In the storage systems described in this application, a range may be a logical concept representing a very sustainable storage unit that can be combined with other ranges (connected or split) to represent a volume. Each scope can be made sustainable by members in a single protection group. The range may provide an LSN-type read / write interface in a contiguous byte sub-range with a fixed size defined at the time of creation. Read / write operations up to a certain extent may be mapped to one or more appropriate segment read / write operations by the containing protection group. As used in this application, the term "volume range" may refer to a range used to represent bytes of a particular sub-range of bytes in a volume.

As mentioned above, volumes, each presented by a protection group, constitute one or more segments, can be composed of a plurality of ranges. In some embodiments, log records indicated in different ranges may have interleaved LSNs. For changes to volumes that are sustainable up to a particular LSN, it may be necessary for all log records up to that LSN to be sustainable, regardless of the extent to which they belong. In some embodiments, the client can track unprocessed log records that are not yet sustainable, and when all ULRs up to a particular LSN become sustainable, it is possible to add a volume sustainable LSN (VDL) message. The VDL can be written to all synchronous mirror segments for the protection group. This is sometimes referred to as "unconditional VDL" and it may be periodically persisted to various segments (or more specifically, to various protection groups) with recording activity taking place on the segments. In some embodiments, the unconditional VDL may be stored in log sector headers.

In various embodiments, operations that may be performed on a segment may include writing an DULR or AULR received from a client (which may involve an operation to update a user page table after writing a DULR or AULR to a tail of a hot- ), An operation to read a cold user page (which may involve locating the exact location of data sectors of a user page and returning them without having to apply any additional DULRs), reading hot user pages Replace the DULRs with AULRs (which may involve applying an operation to find the exact position of the data sectors of the most recent AULR for the page and applying any subsequent DULRs to the user page before returning) Lt; RTI ID = 0.0 > AULR < / RTI > Which may involve), and the like operation of the operation log record. As described in the present application, merging is a process of applying DULRs to a previous version of a user page to create a later version of a user page. Merging user pages may help to reduce read latency since all recorded DULRs may not need to be read and applied on demand before merging (until another DULR is written). It can also help reclaim storage space by making old AULRs and DULRs obsolete (unless there are any snapshots that require log records to exist). In some embodiments, the merge operation may include locating the exact location of the most recent AULR and applying any subsequent DULRs in the sequence without skipping any of the DULRs. As mentioned above, in some embodiments, merging may not be performed within the hot log zone. Instead, it can be performed in a cold log zone. In some embodiments, the merge may also be performed as log records are copied from the hot log zone to the cold log zone.

In some embodiments, the decision to merge the user pages is made by the size of the pending DULR connection to the page (e.g., the length of the DULR connection is determined by system-wide, application-specific, or client- Or exceeds a predetermined threshold for a merge operation), or triggered by a user page read by the client.

FIG. 7 is a block diagram illustrating an exemplary configuration of a database volume 710, in accordance with one embodiment. In this example, data corresponding to each of the various address ranges 715 (shown as address ranges 715a through 715e) is stored in different segments 745 (shown as segments 745a through 745n) do. More specifically, the data corresponding to each of the various address ranges 715 can be structured into different ranges (shown as ranges 725a and 725b, and ranges 735a through 735h), and these ranges The various ranges may be included in different protection groups 730 (shown as striped set 720a and stripe set 720b) (shown as 730a through 730f) without being split or broken apart. In this example, protection group 1 illustrates the use of erasure coding. In this example, protection groups 2 and 3 and protection groups 6 and 7 represent each other's mirrored data sets, and protection group 4 represents a single-instance (non-redundant) data set. In this example, protection group 8 represents a multi-tier protection group combining different protection groups (e. G., It may represent a multi-region protection group). In this example, stripe set 1 720a and stripe set 2 720b illustrate how ranges (e.g., ranges 725a and 725b), in some embodiments, can be split into volumes.

More specifically, in this example, protection group 1 730a includes ranges a through c (735a through 735c), which include ranges 1 through 3 (715a through 715c), respectively, 1 to 4 (745a to 745d). Protection group 2 730b includes a range d 735d, which includes data that is divided from range 4 715d, which is mapped to segments 5 through 7 (745e through 745g). Similarly, protection group 3 730c includes range e 735e, which includes data that is divided from range 4 715d and is mapped to segments 8 and 9 (745h and 745i); Protection group 4 730d includes a range f (735f), which includes data that is divided from range 4 (715d) and is mapped to segment 10 (745j). In this example, protection group 6 730e includes a range g (735g), which includes data that is divided from range 5 (715e) and is mapped to segments 11 and 12 (745k through 745l); Protection group 7 730f includes range h 735h which also includes data that is divided from range 5 715e and is mapped to segments 13 and 14 745m and 745n.

Referring now to FIG. 8, in various embodiments, as described above, a database system may be configured to respond to various access requests (e.g., write requests) for data stored in data pages on storage nodes To generate redo log records and to transmit redo log records to storage nodes storing respective data pages from which redo log records have been generated. The storage nodes may detect a merge event for a particular data page and in response may perform a merge operation on a particular data page. A typical database system, by contrast, can apply checkpoints to the entire system, which flushes all of the generated redo logs to be applied every periodic intervals, to the stored data, And the processing of other tasks.

The method of FIG. 8 may be performed by various components of the log-structured storage system, such as distributed database-optimized storage system 410 (e.g., storage system server node (s) 430, 440, While the method may be described as being performed, the method need not be performed by any particular element in some instances. For example, in some cases, the method of Figure 8 may be performed by some other component or computer system, in accordance with some embodiments. Alternatively, in some cases, the components of the database system 400 may be combined or exist in a manner different than that shown in the example of FIG. In various embodiments, the method of FIG. 8 may be performed by one or more computers of a distributed database-optimized storage system, one of which is illustrated by the computer system of FIG. The method of FIG. 8 is shown as one exemplary implementation of a method for avoiding checkpoints on the entire system. In other implementations, the method of FIG. 8 may include additional blocks in the illustrated blocks or fewer blocks than the illustrated blocks.

As indicated at 810, redo log records linked to a particular data page stored for the database may be maintained. These redo log records (sometimes referred to as URLs as described above) may describe changes to user data. Redo log records can be linked to a specific portion of user data, such as a data page. For example, in some embodiments, redo log records may form successive redo log records that are ultimately linked to a particular data page, with each redo log record pointing to a previously received redo log record for a data page . Using this example, if three redo log records are linked to a particular data page, the most recently received redo log record, which indicates the most recently saved state of the data page, will point to the next most recently received redo log record , Which will eventually point to the third most recently received redo log record. Note that the logical ordering of redo log records represented by each pointer to a previous redo log record does not imply that such redo log records are physically stored in that order. As discussed above with respect to FIG. 6, these redo log records may, in some embodiments, be interleaved with other redo log records linked to other portions of the user data. Accordingly, the preceding examples are not intended to be limiting.

In various embodiments, redo log records may be received from a database system, such as database engine head node 420, which may be stored in a storage node, such as storage node 430, 440, 450, etc. You can manage more than one database. However, in at least some embodiments, the storage node may receive redo log records from one or more additional database systems or nodes where the storage node stores data. These other database systems or nodes may also transmit redo log records linked to specific parts of the data stored for their respective databases at the storage node.

In some embodiments, subsequently received redo log records may be stored. 6 illustrates various embodiments of a method by which such redo log records may be received, processed, and stored at a storage node. Various types of metadata, such as the number or count of redo log records linked to a particular piece of data, such as a data page, may be maintained for the stored redo log records. For example, as in the example given above, if three redo log records are linked to a particular data page, the redo log record count for a particular data page may be maintained at three. Other metadata about the redo log records, such as size or physical location, and portions of the data to which they are linked, may include pointers to various other redo log records or pointers to the most recently saved state of the data page, , ≪ / RTI >

Updates to the metadata maintained for the stored redo log records may be made by making changes to the redo log records themselves, by making changes to specific data pages to which they are linked, or by utilizing redo log records, ≪ / RTI > or other methods or techniques. For example, if a merge operation is performed that applies one or more redo log records linked to a particular data page to generate the current state of the data page, as shown at 830, the redo log record count may be a redo log record for a particular data page May be updated to remove the applied redo log records from the log record count.

In various embodiments, the merge event for a particular data page may be detected, at least in part, as indicated at 820, based on one or more redo log records linked to a particular data page. The detected merge event may indicate that the merge operation can be performed on a particular data page. In at least some embodiments, detecting a merge event for a merge event may occur independently of or in conjunction with the merge events detected for other data pages. Consider a scenario where a particular data page may be a "hot" data page on which many redo log records are received. Redo log records may be received rarely differently for different data pages. Detecting the merge event may be based on the number of redo log records linked to each data page exceeding the merge threshold so that in this scenario the merge event is more specific to the " hot "data Pages can be detected more frequently.

Detecting a merge event may be accomplished by monitoring a component or process that is capable of executing as a background process that foreground processes that process read, write, and other access requests can be performed (or delayed) prior to detection of the merge event May be performed as part of the node. Detection of the merge event may occur at non-periodic intervals, such as when the workload of the periodic or storage node is less than the workload threshold.

At least in part, various methods and techniques for merging events based on redo log records linked to a particular data page may be implemented. For example, in at least some embodiments, the merge threshold may be utilized to detect merge events. The merge threshold may define the number of redo log records that can be linked to a particular data page before the merge event is detected. For example, if a particular data page has eleven redo log records that exceed the merge threshold of ten redo log records, a merge event can be detected. Different merge thresholds may be utilized for different data pages. For example, consider a "hot" data page scenario that receives frequent redo log records linked to data pages. A higher merge threshold may be utilized for a "hot" data page than a data page receiving redo log records less frequently, thereby reducing the number of merge operations performed on "hot" . Alternatively, in some embodiments, the same or similar merge thresholds may be utilized. The merge threshold may also be combined with other techniques or components. For example, using other components to calculate when the merge threshold is likely to be exceeded and indicating to the background monitor or other process performing merge event detection that the redo log record count for a particular data page should be reviewed Making timers or other components for.

In at least some embodiments, a merge threshold for a particular data page may be determined (or for a particular set of data pages). For example, in some embodiments, the merge threshold may be determined according to a user-defined merge threshold. The user-defined merge threshold may be a merge threshold that is requested, determined, or displayed from the database system to the storage node, such as the database engine head node 420, or may be a merge threshold to be used by the client of the database system to detect the merge event . In some embodiments, the merge threshold may be determined based on workload or performance of the storage node. For example, in some embodiments, if the degree of workload or performance indicates that the ability to perform the merge operations is low, then the merge threshold may be such that the number of merge events detected is to be processed by the storage node by its current workload Can be increased. In some embodiments, the rate or frequency at which redo log records are received for a particular data page may be computed and used to determine the merge threshold. In at least some embodiments, the size of the redo log records, the location of the redo log records in the physical storage, the space available to store the redo log records, and / Various other characteristics may be used to determine the merge threshold, such as the time that may be performed to apply to the version.

In response to detecting a merge event for a particular data page, one or more redo log records linked to a particular data page may be migrated to a specific data page in order to create a particular data page in its current state, Lt; / RTI > In at least some embodiments, applying a redo log record linked to a particular data page is performed as part of a merge operation. Merge operations or merging as described above may apply redo log records, such as DULRs, to the previous version of the user page to create a later version of the user page. In some embodiments, the merge operation may include locating the most recent AULR (e.g., a previously stored version of the data page) and applying any subsequent DULRs sequentially without skipping any of the DULRs . For example, if three DULRs are received and linked to an AULR, the first received DULR is applied to the AULR (thus applying the first received change on the previously stored data page). The next received DULR is then applied, and finally the most recent DULR is applied to apply the DULRs in a sequence determined based on the reception of the DULR at the storage node. In some embodiments, a new AULR is generated as the current state of a particular data page. The metadata discussed above, such as the redo log record count, is updated to reflect the application of redo log records, and for the redo log record, the number may be removed from the count.

In at least some embodiments, a delay may occur or may be implemented between detecting the merge event, as indicated at 820, and applying redo log records, as indicated at 830. For example, the workload of the storage node performing the detecting and applying may determine the delay between performing the applying redo log records and detecting the merge event. Similarly, applying redo log records in response to detecting a merge event may be performed only when not performing foreground processes, such as processing various access requests (e.g., read requests or write requests) Or as part of a background process, which is performed or reduced. The application of redo logs to delayed merge operations or data pages may be performed using data such as priority queues or first in first out (FIFO) queues that determine the order, sequence, or timing when data pages should apply redo log records Lt; / RTI > For example, if a "hot" data page has a detected merge event, as in the scenario described above, it may be more efficient to perform the application of redo logs to a "hot" data page instead of another data page. As a result of performing or delaying the application of redo log records as a background process, one or more additional redo log records linked to the data page in which the merge event was detected may be received. In at least some embodiments, these additional redo log records may be applied when other redo log records are applied to a previously stored version of the data page.

As illustrated in FIG. 4, multiple storage nodes 430, 440, 450, etc. may be implemented as part of the distributed storage service. The various methods and techniques described above with respect to FIG. 8 above may be performed by these multiple storage nodes independently of each other. Each storage node may perform detection of merge events and applying one or more redo log records in response to the same or different times as well as determining different or the same merge thresholds.

Reference is now made to FIG. 9A, which illustrates a series of examples illustrating a method for performing fast failover for a distributed database system, in accordance with some embodiments. Failover in a typical database system is a difficult process. In these typical systems, upon recovery from a database system failure, a clean version of the database is obtained, and then a significant recovery time is generated before the database can be accessed, so that all redo logs from transactions not stored on disk The record must be replayed to recover the database to its current state prior to the database system failure. Figure 9A, by contrast, provides examples of fast failover for a distributed database system that can provide faster and more efficient techniques for performing failover.

The database client 906, such as the client 250 described above with respect to FIG. 2, in the scene 992 may be coupled to a database head node, such as the database head node 430 described above with respect to FIG. 4, Lt; RTI ID = 0.0 > 902, < / RTI > Storage nodes 908 may be one or more storage nodes that implement a log-structured data storage device for a database implemented by database head node 902. Various access requests may be received and subsequently serviced by the database head node 902 in retrieving the accessed data from the storage nodes 908. [ Redo log records, such as the redo log records described above with respect to FIG. 8, may be generated and transmitted to storage nodes 908 instead of transmitting user data. Redo log records may be maintained at storage nodes 908. [ In at least some embodiments, the merge operation may be performed in response to the detection of a merge event, as described above with respect to Fig.

The scene 994 illustrates the failure of the database head node 902. A database head node failure can be any type of system failure that keeps the database head node from continuing to function, such as power loss, memory unavailability, system glitches, and the like. Any communication between the database client 906 and the database head node 902 may or may not be transmitted, as indicated in the diagram. Thus, no access to the database may be provided. Similarly, no communication between the storage nodes 908 and the database head node 902 may be transmitted or received, and thus no requests for data stored against the database may be processed.

In scene 996, a restore operation may be illustrated. A new database head node 904, which may be a version of a head node application program that is restarted on the same hardware or another instance of a head node that is started on different hardware, may be brought online. Connections with storage nodes 908 may be established by database head node 904, as illustrated. Scenes 998 show that, in establishing connections with storage nodes 908, the same database as that implemented at database head node 902 can be made available for access at new database head node 904 . Access requests, such as read requests or write requests, may be transmitted from the database client 906 to the new database head node 904 via the network 260. Since these redo log records have already been sent to the storage nodes 908 that can provide the current version of the data stored for the database to the new database head node 908 to service the access request, ) May not need to replay redo log records to obtain the current state of the data prior to the database head node failure. Storage nodes 908 may apply redo log records to a previously stored version of particular data when a request for specific data is received. Alternatively, the current state of the particular data may be pre-stored in storage nodes having any redo log records pointed to a specific already applied, such as when a merge event is detected as described above with respect to FIG. 8 have.

Figure 9B is a flow chart illustrating a method for performing fast failover for a distributed database system, in accordance with some embodiments. In various embodiments, a database head node failure may occur. This head node failure can prevent any form of communication, modifications, or access to the database being implemented and managed by the failed database head node. For example, a database system client, such as database client 250 illustrated in FIG. 2, may not be able to send read or write requests to a failed database head node. The failure of the database head node may be detected, such as by the web service platform 200 described above in FIG. 2, or some other system or component. In response to a failure of the head node, a restarted database head node or a new database head node (e.g., a new database head node instance hosted with one of the same or different hardware as the previously failed head node) . In some embodiments, this recovery operation may include various elements shown in FIG. 9B, but it is not limited to these elements.

Recovery from the database head node failure may occur, as indicated at 910. [ Recovery can be performed and determined to be completed in various ways. For example, a database head node application may have various states when it is ready to perform, such as performing various tests, enabling various devices, and the like. As part of this process, the ready state can be determined for a database head node that can indicate the completion of recovery from a node failure. Upon recovery from a database node failure, as indicated at 910, a connection with one or more storage nodes that store data for the database may be established, as indicated at 920. [

As described above with respect to FIG. 9A and the various other figures, the database may be implemented and managed by a database head node, such as the database head node 320 or 440 described in FIGS. 3 and 4. FIG. As part of implementing the database, access requests, such as the read request write requests described above, may be processed at the database head node. Redo log records reflecting changes to the database in at least some embodiments may be transmitted to one or more storage nodes, such as the storage nodes 450 described above in FIG. 4, which are stored in storage nodes It reflects changes to the data. Storage nodes that store data to be modified, such as specific data pages or other portions of data, may receive redo log records that are linked to portions of data, such as data pages, that must be altered. These redo log records may then be updated in response to requests for the current version of the data page, or in response to detecting such a merge event, at some other time, on a previously stored version of a portion of the data, (E. G., A merge operation). Since the redo log records for the database are maintained for the database implemented at the database head node, such as the various methods discussed above, the storage nodes are guaranteed, in some embodiments, up to the time of the database head node failure The current state of the data to be transmitted to the database head node.

Storage nodes establishing connections can be identified. For example, the client-side storage service driver 425 described above in FIG. 4 may store information indicating which storage nodes store data for the database and which portions of the database are stored at storage nodes . A connection request, or some other communication message may be transmitted using one of the various communication methods discussed above with respect to FIG. Similarly, acknowledgments and other information about the status of the storage node and / or the database head node may be exchanged.

As indicated at 920, upon establishing a connection with one or more storage nodes, the database may be made available for access, as indicated at 930. In some embodiments, an access may be provided for one or more access requests (e.g., read requests, write requests). An indication of the availability of the database may be generated and transmitted to the client. For example, a message can be sent to the database client that the database is available for access. Such a message may be sent via the web service platform 200, or some other communication platform or device, as illustrated in FIG. As noted above, in typical database systems, a replay of redo log records must be performed before making the database available. However, in at least some embodiments, the database may be made available without replaying the redo log records. Note that the term "replay" when used with redo log records generally means applying one or more redo log records to a previously stored version of the data.

In at least some embodiments, the storage node may be able to detect or otherwise recognize a database head node failure. In response to detecting a database head node failure, the storage node may perform a truncation operation on the redo log records received at the storage node. The truncation operation may determine or identify redo log records that are part of a system transaction that did not complete before the failure of the database head node. These identified redo log records may then be removed or otherwise marked, moved, or otherwise identified such that they may not be applied to the linked data pages. For example, if the store page holds 5 redo log records for a particular data page and the last 3 redo log records are part of a system transaction that did not complete before the database head node failure, By applying an old redo log record, you can ignore the three most recent redo log records for a data page when generating the current state of the data page. In at least some embodiments, the truncation operation may be performed on the storage node with the affected redo log records before enabling the established connection with the restored database head node. The database engine head node may, in some embodiments, determine or identify redo log records that are part of a system transaction that did not complete before the failure of the database head node, and that these identified redo log records indicate that they are part of the linked data pages May be configured to send a notification to the storage nodes that they may be removed or otherwise marked, moved, or otherwise identified as not being applicable. For example, a client-side storage service driver, such as the client-side storage service driver 325 described above with respect to FIG. 3, may perform the techniques described previously. These techniques for describing the cutting operation may, in some embodiments, be performed as part of a background process.

In at least some embodiments, a system transaction may be an operation for performing or implementing a user transaction, or another form of operation or operations. A user transaction may include a number of system transactions for performing various operations or operations from a received access request. For example, an insert command to the database may be received. As a user transaction, this insert command may include a number of system transactions to perform insertions, such as to interact with database data structures, e.g., b -trees, to perform inserts. In at least some embodiments, an incomplete user transaction is a user transaction that may not have all system transactions included in the user transaction completed (not sustainable). Similarly, a system transaction may also be uncompleted. Redo log records reflecting changes made to data stored for a database as part of user and system transactions may, in some embodiments, be identified as specific user and / or system transactions.

9C is a flow chart illustrating a method for processing access requests in a restored database, in accordance with some embodiments. As mentioned above, in at least some embodiments, access requests may be received at the database head node that made the database available for access. The access requests may be a read request, a write request, or any other request to acquire or modify data stored for the database. As illustrated in FIG. 9C, an access request may be received for the database, as indicated at 940. In response, a request for one or more data pages from one or more storage nodes may be made, as indicated at 950 (both access requests from clients and data requests from the database head node are shown in FIG. 5 Is covered in more detail. The current state of the one or more requested data pages may be received from the storage nodes, as indicated at 960. As discussed above, this current state may be generated by replaying or applying previously received redo log records to a previously stored version of the data page, or by returning a previously stored version of the current data page . In various embodiments, each data page or a portion of the requested data may be determined, generated, and / or transmitted back to (e.g., in a slow manner) its current state in response to receiving a request for data Lt; / RTI >

In at least some embodiments, Undo log records may be maintained at the database head node. As discussed above, Undo log records can record changes to be applied to data stored against the database to cancel changes made to the data, such as in an event of an incomplete user transaction. A user transaction may include one or more redo log records and undo log records, including multiple changes (such as multiple system transactions) to data stored against the database. A user transaction can be undone when all changes to a user transaction are not committed (persisted). Such as the transaction log 340 described above with respect to FIG. 3, that the transaction tables are not committed before the database head node failure and that their associated portions of the data stored in the storage nodes are incomplete May be implemented for display. As indicated at 970, a determination may be made as to whether the received data page is affected by an incomplete user transaction, as indicated by the transaction table. If so, as indicated by the exit of affirmation, one or more of the undo log records may be applied to the data page to cancel the changes made by the incomplete transaction, as shown at 972, and generate a new current state of the data page have. After the undo log records are applied, or after the data page is not affected by an incomplete user transaction, the current state of the data page may be provided to service the access request, as indicated at 980. [

In at least some embodiments, a background process may be performed, based on the transaction table, to determine or identify portions of data that are affected by an incomplete user transaction. Requests for the current state of portions of data, such as data pages, that are affected by incomplete user transactions can be sent and received. The undo log records may then be applied to cancel the changes indicated to these data pages by the incomplete user transaction. In various embodiments, the database cache may be updated with these data pages after the undo log records are applied.

In at least some embodiments, a previously recorded snapshot may be used to restore the state of the database to a previous state. For example, prior to making the database available for access, a request to be restored to a state corresponding to a previously recorded snapshot of data for the database may be sent to the storage nodes. The snapshot can be a timestamp or other marker or indication for redo logs stored on storage nodes that allow previously received redo log records to be replayed to the snapshot point (e.g., a timestamp or marker) Wherein the recovery includes applying one or more of the plurality of redo logs to a previous version of the data. Additional discussion of implementing snapshots on storage nodes is provided above.

Although the methods and techniques of FIGS. 9B and 9C have been described as being performed by the various components of the database system, such as database engine head node 420, the method may be performed by any particular component in some cases It does not need to be. For example, in some cases, the methods of Figures 9B and 9C may be performed by some other component or computer system, in accordance with some embodiments. Alternatively, in some cases, the components of the database system 400 may be combined or exist in a manner different than that shown in the example of FIG. In various embodiments, the methods of FIGS. 9B and 9C may be performed by one or more computers of a distributed database system, one of which is illustrated by the computer system of FIG. The methods of Figures 9B and 9C are shown as exemplary implementations of methods for fast failback recovery of a distributed database system. In other implementations, the methods of Figures 9B and 9C may include additional blocks or fewer blocks than shown.

The methods described in this application may be implemented by any combination of hardware and software in various embodiments. For example, in one embodiment, the methods may be performed on a computer system (e.g., a computer system as in FIG. 10) that includes one or more processors executing program instructions stored on a computer readable storage medium coupled to the processors, Lt; / RTI > Program instructions may be implemented to implement the functionality described in the present application (e.g., the functionality of the various servers and other components that implement the database services / systems and / or storage services / system described in this application) .

10 is a block diagram illustrating a computer system configured to implement at least a portion of the database systems described in this application, in accordance with various embodiments. For example, computer system 1000 may be, in different embodiments, either one of a plurality of storage nodes of an individual distributed database-optimized storage system that stores databases and associated metadata on behalf of clients of a database tier, Lt; RTI ID = 0.0 > node < / RTI > The computer system 1000 includes, but is not limited to, a personal computer system, a desktop computer, a laptop or notebook computer, a mainframe computer system, a handheld computer, a workstation, a network computer, a customer device, an application server, , A mobile phone, or any type of device, generally including any type of computing device.

The computer system 1000 may include one or more processors 1010 (each of which may include multiple cores, which may be a single or a plurality of cores) coupled to the system memory 1020 via an input / output (I / O) Multi-threaded). The computer system 1000 further includes a network interface 1040 coupled to the I / O interface 1030. In various embodiments, computer system 1000 includes a single processor system including one processor 1010, or some (e.g., 2, 4, 8, or another suitable number) processor 1010 Lt; / RTI > system. Processors 1010 may be any suitable processors capable of performing existing instructions. For example, in various embodiments, the processors 1010 may be any architecture of various instruction set architectures (ISAs), such as x86, PowerPC, SPARC, or MIPS ISAs, or any other suitable ISA Lt; RTI ID = 0.0 > and / or < / RTI > In multiple processor systems, each of the processors 1010 may, but need not necessarily, implement the same ISA. Computer system 1000 also includes one or more network communication devices (e.g., network interface 1040) for communicating with other systems and / or components via a communication network (e.g., Internet, LAN, . For example, a client application executing on the system 1000 may include a network interface 1040 to communicate with a server application on a cluster or on a single server that implements one or more of the components of the database systems described in this application. Can be used. In another example, an instance of a server application executing on computer system 1000 may be accessed by another instance of a server application (or another server application) that may be implemented on other computer systems (e.g., computer systems 1090) Lt; RTI ID = 0.0 > 1040 < / RTI >

In the illustrated embodiment, the computer system 1000 also includes one or more persistent storage devices 1060 and / or one or more I / O devices 1080. In various embodiments, persistent storage devices 1060 may correspond to disk drives, tape drivers, solid state memory, other mass storage devices, or any other persistent storage device. The computer system 1000 (or an operating system running on a distributed application or on it) may store instructions and / or data on persistent storage devices 1060 on demand and may retrieve stored commands and / . For example, in some embodiments, the computer system 1000 may host a storage system server node, and the persistent storage 1060 may include SSDs connected to the server node.

Computer system 1000 includes one or more system memory 1020 configured to store instructions and data accessible by processor (s) In various embodiments, the system memories 1020 may comprise any suitable memory technology, such as, for example, one or more caches, static random access memory (SRAM), DRAM, RDRAM, EDO RAM, DDR 10 RAM, (SDRAM), Rambus RAM, EEPROM, non-volatile / flash-type memory or any other type of memory). The system memory 1020 may include program instructions 1025 executable by the processor (s) 1010 to implement the methods and techniques described in this application. In various embodiments, program instructions 1025 may be stored in any interpreted language such as platform-specific binary, JavaTM byte-code, or in any other language, such as C / C ++, Java TM, Lt; / RTI > For example, in the illustrated embodiment, the program instructions 1025 may, in different embodiments, include a plurality of stores (e.g., databases) of individual distributed database-optimized storage systems that store databases and associated metadata on behalf of clients of a database tier One of the nodes, or a database engine head node of the database tier. In some embodiments, program instructions 1025 may implement a number of individual clients, server nodes, and / or other components.

In some embodiments, program instructions 1025 may be executable instructions (not shown) to implement an operating system (not shown), which may be any of a variety of operating systems, such as UNIX, LINUX, SolarisTM, MacOSTM, Lt; / RTI > Any or all of the program instructions 1025 may be provided as a computer program product or software that may include non-transitory computer readable storage media having instructions stored thereon, May be used to program a computer system (or other electronic devices) to perform a process in accordance with embodiments. Non-transient computer readable storage media may include any mechanism for storing information in a form (e.g., software, processing application) readable by a machine (e.g., a computer). Generally speaking, the non-transitory computer accessible medium may be a memory medium such as a disk or DVD / CD-ROM coupled to the computer system 1000 via magnetic or optical media, e.g., I / O interface 1030, Computer-readable storage media. Non-volatile computer readable storage media may also include any volatile or non-volatile media such as RAM (e.g., SDRAM, DDR SDRAM, RDRAM, SRAM, etc.), ROM, May be included in some embodiments of computer system 1000 as other types of memory. In other embodiments, the program instructions may be transmitted over a network (not shown), such as a network (e.g., a cellular telephone, a cellular telephone, a cellular telephone, a cellular telephone, Signals, digital signals, and the like) in the form of digital signals.

In some embodiments, system memory 1020 may include a data store 1045, which may be configured as described in this application. (E.g., on a database engine head node), such as, for example, transaction logs, undo logs, cached page data, or other information used in performing the functions of the database tiers described in this application The information described in this application may be stored in the persistent store 1060 in various instances and in various embodiments in the data store 1045 or in another portion of the system memory 1020 on one or more nodes , And / or one or more remote storage devices (1070). Similarly, the use of the information described in this application (e.g., redo log records, merged data pages, and / or to perform the functions of the distributed storage systems described in this application) as stored by the storage tier May be stored in various instances and in various embodiments in the data store 1045 or in another portion of the system memory 1020 on one or more nodes to the persistent store 1060 and / May be stored in the storage device 1070. In general, system memory 1020 (e.g., data storage 1045 in system memory 1020), persistent storage 1060, and / or remote storage 1070 may include data blocks, , Metadata associated with the data blocks and / or their state, database configuration information, and / or any other information available for implementing the methods and techniques described in this application.

In one embodiment, the I / O interface 1030 is coupled to the processor 1010, the system memory 1020 and any other devices in the system, such as through the network interface 1040 or other peripheral interfaces, / O < / RTI > traffic. In some embodiments, the I / O interface 1030 is used by the other component (e.g., processor 1010) to transfer data signals from one component (e.g., system memory 1020) Timing, or other data transformations to transform the data into a format suitable for use with the present invention. In some embodiments, the I / O interface 1030 is a device that is connected via various types of peripheral buses, such as, for example, various peripheral component interconnect (PCI) bus standards or Universal Serial Bus Lt; RTI ID = 0.0 > and / or < / RTI > In some embodiments, the functionality of the I / O interface 1030 may be distributed to two or more discrete components, such as, for example, a north bridge and a south bridge. Also, in some embodiments, some or all of the functionality of the I / O interface 1030, such as an interface to the system memory 1020, may be included directly in the processor 1010.

The network interface 1040 can be any type of network interface capable of communicating data such as, for example, computer system 1000 and other computer systems 1090 (including one or more storage system server nodes, database engine head nodes, and / Or other devices that are connected to the network, such as < / RTI > Network interface 1040 may also be configured to allow communication between computer system 1000 and various I / O devices 1050 and / or storage 1070. The input / output devices 1050 may, in some embodiments, accept or reject data by one or more display terminals, a keyboard, a keypad, a touchpad, a scanning device, a voice or optical recognition device, But may include any other device suitable for retrieval. A plurality of input / output devices 1050 may reside in the computer system 1000 or may be distributed over various nodes of the distributed system including the computer system 1000. In some embodiments, similar input / output devices may be separate from the computer system 1000 and may be coupled to one of the distributed systems including the computer system 1000 via a wired or wireless connection, such as via the network interface 1040 Can interact with the above nodes. Network interface 1040 may typically support one or more wireless networking protocols (e.g., Wi-Fi / IEEE 802.11, or another wireless networking standard). However, in various embodiments, the network interface 1040 may support communication over any suitable wired or wireless common data networks, such as, for example, other types of Ethernet networks. In addition, the network interface 1040 may be implemented over telephony / telephone networks, such as analog voice networks or digital fiber communication networks, over a storage area network, such as Fiber Channel SANs, or any other suitable type of network and / Lt; RTI ID = 0.0 > and / or < / RTI > In various embodiments, the computer system 1000 may include more, fewer, or different components (e.g., displays, video, cards, audio cards, peripheral devices, ATMs Interface, an Ethernet interface, a Frame Relay interface, and the like).

It is noted that any of the embodiments of distributed system embodiments described in this application, or any of the components thereof, may be implemented as one or more web services. For example, a database engine head node in a database tier of a database system may provide other types of data storage services employing the distributed storage system described in this application for clients as database services and / or web services. In some embodiments, the web service may be implemented by software and / or hardware systems designed to support co-operable machine-to-machine interaction over a network. A Web service may have an interface that is described in a machine processable format, such as the Web Services Description Language (WSDL). Other systems may interact with the web service in a manner prescribed by the description of the interface of the web service. For example, a web service may define various operations that other systems may apply, and may define a specific application programming interface (API) that other systems may expect to follow when requesting various operations.

In various embodiments, the web service may be requested or applied through the use of a message containing parameters and / or data associated with the web service request. Such messages may be formatted and / or encapsulated using a protocol such as Simple Object Access Protocol (SOAP) according to a particular markup language, such as Extensible Markup Language (XML). To perform a web service request, the web service client assembles a message containing the request and sends the message to an addressing application corresponding to the web service using an Internet-based application layer transport protocol, such as Hypertext Transfer Protocol (HTTP) To a possible endpoint (e.g., a unique resource location indicator (URL)).

In some embodiments, web services may be implemented using "RESTful" (Representational State Transfer) techniques rather than message based technologies. For example, a web service implemented in accordance with RESTful technology can be applied through parameters contained in HTTP methods such as PUT, GET, or DELETE rather than encapsulated within a SOAP message.

The foregoing can be better understood by considering the following sections:

One. As a system:

At least one database engine head node of a database service:

A method for generating one or more redo log records linked to a specific data page among a plurality of data pages stored in a storage node among a plurality of storage nodes implementing a distributed storage service, the storage nodes comprising a plurality of data pages including a specific data page Generate one or more redo log records, wherein each of the one or more redo log records is generated in response to one or more access requests for data stored in a particular data page;

At least one database engine head node of the database service, configured to transmit one or more redo log records to a storage node;

A storage node of a distributed storage service comprising:

Storing one or more received redo log records linked to a particular data page;

Determine that one or more redo log records linked to a particular data page exceeds a merge threshold;

Wherein the merging operation comprises applying one or more redo log records linked to a data page specific to a previously stored version of a particular data page to generate a particular data page in its current state, And a storage node of the distributed storage service configured to perform the merge operation.

2. In Section 1,

The storage node is further configured to maintain a redo log record count for each of the plurality of data pages;

To determine that one or more of the redo log records that are linked to a particular data page exceeds the merge threshold, wherein the at least one of the redo log records is maintained for a particular data page to determine that the redo log record count exceeds the merge threshold System.

3. The system of clause 2, wherein the storage node is further configured to, in response to performing the merge operation, update the redo log record count to remove one or more redo log records linked to a particular data page from the redo log record count.

4. In Section 1,

The database engine head node previously created and transmitted one or more other redo log pages linked to different ones of the plurality of data pages stored in the storage node;

The database engine head node is:

Upon retrieval from a database engine head node failure, to send a request for a current state of a data page specific to a storage node and a current state of another data page;

The storage node is:

Receive a request for a current state of a particular data page and a current state of another data page from a database engine head node;

Send a previously generated current state of a particular data page to a database engine head node in response to receiving a request for a particular data page;

In response to receiving a request for another data page:

Performing a merge operation that applies one or more other redo log records linked to different data pages to a previously stored version of another data page to generate another data page in its current state;

And send the current state of another data page to the database engine head node.

5. A method comprising:

By one or more computing devices:

Maintaining one or more redo log records linked to a particular data page stored for the database;

Detecting a merge event for a particular data page based, at least in part, on one or more redo log records linked to a particular data page; And

In response to detecting a merge event for a particular data page, applying one or more redo log records linked to a data page specific to a previously stored version of a particular data page to generate a particular data page in its current state The method comprising:

6. 5. The method of clause 5, wherein detecting the merge event for a particular data page occurs non-periodically.

7. 5. The method of clause 5, wherein detecting the merge event for a particular data page further comprises determining that one or more redo log records linked to a particular data page exceeds a merge threshold.

8. 7. The method of clause 7, wherein detecting the merge event for a particular data page further comprises determining a merge threshold according to a user-defined merge threshold.

9. In Section 7,

Maintaining one or more additional redo log records linked to different data pages stored for the database; And

Further comprising determining that one or more additional redo log records linked to different data pages exceed different merge thresholds, different from the merge threshold.

10. In clause 5, one or more computing devices together implement a storage node of a plurality of storage nodes implementing a distributed storage service, wherein one or more redo log records include a specific data page stored for a database across a plurality of storage nodes Wherein the plurality of redo log records are included in a plurality of redo log records each linked to one of the plurality of data pages, and a plurality of redo log records are received from the database system.

11. 9. The method of clause 10 wherein at least one of the plurality of storage nodes is capable of holding, detecting, and applying different data pages of a plurality of data pages stored in one or more other storage nodes Performed.

12. 12. The method of clause 12, wherein the detecting and applying are performed at different times for different ones of the data pages stored in one or more other storage nodes.

13. In clause 5, one or more computing devices implement storage nodes among a plurality of storage nodes implementing distributed storage services,

In response to detecting a merge event for a particular data page:

Performed at least in part by the storage node, before applying one or more redo log records linked to a data page specific to a previously stored version of a particular data page to create a particular data page in its current state, Further comprising delaying the start of applying the one or more redo log records based on the workload of the processes.

14. 20. A non-transient, computer readable storage medium having stored thereon program instructions, the program instructions being executed by one or more computing devices,

Maintaining one or more redo log records linked to a particular data page stored for the database;

Determining that one or more redo log records linked to a particular data page exceeds a merge threshold; And

Computer readable storage medium that implements applying one or more redo log records linked to a data page specific to a previously stored version of a particular data page to produce a particular data page in its current state, .

15. 14. The method of clause 14 wherein one or more computing devices together implement a storage node of a distributed storage service and the program instructions further include determining at least in part the merge threshold based on performance of the storage node when executed by the one or more computing devices Non-volatile, computer readable storage medium embodying the invention.

16. 14. The computer readable medium of claim 14, wherein the program instructions further comprise determining, when executed by the one or more computing devices, at least in part, a merge threshold based on the frequency of received redo log records linked to a particular data page, Computer readable storage medium.

17. 14. The non-transitory, computer readable storage medium of claim 14, wherein the program instructions further implement, when executed by the one or more computing devices, at least in part, determining a merge threshold based on storage availability.

18. 14. The computer readable medium of claim 14, wherein the program instructions further execute, when executed by the one or more computing devices, at least in part, to determine a merge threshold based on a time at which the applying of the one or more redo log records can be performed, Transient, computer readable storage medium.

19. 15. The method of clause 14, wherein the program instructions, when executed by the one or more computing devices, further at least partially implement determining a merge threshold based on one or more sizes of one or more redo log records linked to a particular data page, Transient, computer readable storage medium.

20. 14. The method of clause 14, wherein the program instructions further execute, when executed by the one or more computing devices, to determine a merge threshold based, at least in part, on one or more storage locations for one or more redo log records linked to a particular data page , Non-transient, computer readable storage medium.

21. The method of claim 14, wherein applying one or more redo log records linked to a data page specific to a previously stored version of a particular data page to generate a particular data page in its current state is performed as a background process, - Temporary, computer readable storage medium.

22. 14. The method of clause 14 wherein the program instructions are executed by one or more computing devices: one or more redo log records linked to a data page specific to a previously stored version of a particular data page to generate a particular data page in its current state Further comprising updating a redo log record count for a particular data page to remove one or more redo log records linked to a particular data page from the redo log record count, Possible storage medium.

23. 22. The computer program of claim 22, wherein when the program instructions are executed by one or more computing devices:

Receiving another redo log record linked to a particular data page; And

A non-transient, computer readable storage medium that implements updating a redo log record count for a particular data page to include other redo log records linked to a particular data page.

The various methods illustrated in the Figures and described in the present application represent exemplary embodiments of the methods. The methods may be implemented manually in software, in hardware, or a combination thereof. The order of any method may be varied, and various elements may be added, reordered, combined, omitted, modified, and so on.

While the above embodiments have been described in considerable detail, many changes and modifications may be made as will become apparent to those skilled in the art upon a complete understanding of the disclosure. It is intended that the following claims include all such modifications and alterations, and accordingly, the description is to be regarded as illustrative rather than in a limiting sense.

Claims (15)

At least one database engine head node of a database service:
A method for generating one or more redo log records linked to a specific data page among a plurality of data pages stored in a storage node among a plurality of storage nodes implementing a distributed storage service, Wherein the one or more redo log records are generated in response to one or more access requests for data stored in the particular data page;
At least one database engine head node of the database service, configured to transmit the at least one redo log record to the storage node;
The storage node of the distributed storage service comprising:
Store the received one or more redo log records linked to the specific data page;
Determine that the one or more redo log records linked to the particular data page exceeds a merge threshold;
The merging operation applying the one or more redo log records linked to the particular data page to a previously stored version of the particular data page to generate the particular data page in its current state, Wherein the storage node of the distributed storage service is configured to perform the merging operation. The method according to claim 1,
Wherein the storage node is further configured to maintain a redo log record count for each of the plurality of data pages;
Wherein the at least one output node is maintained for the particular data page to determine that the at least one redo log record linked to the particular data page exceeds the merging threshold, The system further comprising: 3. The method of claim 2, wherein the storage node is responsive to performing the merge operation to update the redo log record count to remove the one or more redo log records linked to the particular data page from the redo log record count ≪ / RTI > The method according to claim 1,
The database engine head node previously generated and transmitted one or more other redo log pages linked to another of the plurality of data pages stored in the storage node;
The database engine head node comprising:
To send a request for the current state of the particular data page and the current state of the other data page to the storage node upon recovery from a database engine head node failure;
The storage node comprising:
Receive the request for the current state of the particular data page and the current state of the other data page from the database engine head node;
In response to receiving the request for the particular data page, sending a previously generated current state of the particular data page to the database engine head node;
In response to receiving the request for the other data page:
Performing a merge operation to apply the one or more other redo log records linked to the different data page to a previously stored version of the other data page to generate the other data page in its current state;
And send the current state of the other data page to the database engine head node. By one or more computing devices:
Maintaining one or more redo log records linked to a particular data page stored for the database;
Detecting a merge event for the particular data page based at least in part on the one or more redo log records linked to the particular data page; And
In response to detecting the merge event for the particular data page, the previously stored version of the particular data page to generate the particular data page in its current state, And applying a redo log record. 6. The method of claim 5, wherein detecting the merge event for the particular data page occurs non-periodically. 6. The method of claim 5, wherein detecting the merge event for the particular data page further comprises determining that the one or more redo log records linked to the particular data page exceeds a merge threshold. 8. The method of claim 7, wherein detecting the merge event for the particular data page further comprises determining the merge threshold according to a user-defined merge threshold. The method of claim 7,
Maintaining one or more additional redo log records linked to different data pages stored for the database; And
Further comprising determining that the one or more additional redo log records linked to the different data pages exceed a different merge threshold different from the merge threshold. 6. The method of claim 5, wherein the one or more computing devices together implement a storage node of a plurality of storage nodes implementing distributed storage services, wherein the one or more redo log records are stored in the storage medium, Wherein the plurality of redo log records are included in a plurality of redo log records each linked to one of a plurality of data pages including a data page, wherein the plurality of redo log records are received from a database system. 11. The method of claim 10, wherein the one or more other storage nodes of the plurality of storage nodes are maintained for different ones of the plurality of data pages stored in the one or more other storage nodes, To perform the steps. 8. The method of claim 7, wherein the detecting and applying are performed at different times for the different ones of the data pages stored in the one or more other storage nodes. The system of claim 5, wherein the one or more computing devices together implement storage nodes among a plurality of storage nodes implementing distributed storage services,
In response to detecting the merge event for the particular data page:
At least in part, prior to applying the one or more redo log records linked to the particular data page to a previously stored version of the particular data page to create the particular data page in its current state, Further comprising delaying the start of applying said one or more redo log records based on the workload of other processes performed by said one or more redo log records. As a system:
One or more processors; And
One or more memory for storing program instructions, wherein the program instructions are executed by one or more processors:
Maintaining one or more redo log records linked to a particular data page stored for the database;
Determining that the one or more redo log records linked to the particular data page exceeds a merge threshold; And
And applying the one or more redo log records linked to the particular data page to a previously stored version of the particular data page to generate the particular data page in its current state. 15. The computer-readable medium of claim 14, wherein the program instructions are executable by the one or more processors to: link to the particular data page to the previously stored version of the particular data page to generate the particular data page in its current state In response to applying the one or more redo log records, updating the redo log record count for the particular data page to remove one or more redo log records linked to the particular data page from the redo log record count ≪ / RTI >

Images